Cover page

Image Name: The Blue Heart

The blue heart symbolizes the urgent need for a harmonious relationship between technology and nature, where

artiﬁcial intelligence can play a pivotal role in environmental conservation. As we harness AI to monitor

ecosystems and predict climate changes, we must ensure that our innovations promote health and sustainability

for all living beings. By embracing this blue heart, we commit to a future where technology enhances our

connection to the planet, fostering both ecological balance and human well-being. Let this symbol inspire us to

create intelligent solutions that protect our environment while nurturing our health.

Managing Editor Chief Editor Editorial Board Correspondence

Ninan Sajeeth Philip Abraham Mulamoottil K Babu Joseph The Chief Editor

Ajit K Kembhavi airis4D

Geetha Paul Thelliyoor - 689544

Arun Kumar Aniyan India

Sindhu G

Journal Publisher Details

Publisher : airis4D, Thelliyoor 689544, India

Website : www.airis4d.com

Email : nsp@airis4d.com

Phone : +919497552476

Editorial

by Fr Dr Abraham Mulamoottil

airis4D, Vol.2, No.9, 2024

www.airis4d.com

The editorial of this edition of airis4D is pleased

to spotlight Chinar Deshpande, a 12th-grade student at

Aditya English Medium School, who is deeply passion-

ate about research in Computer Vision, Machine Learn-

ing, and the intersection of AI and Education. Chinar

has interned at EagleRobotLab in Bangalore, where he

contributed to developing humanoid robots designed

to teach and assist instructors in classrooms. He is

also actively collaborating with the Pune Knowledge

Cluster, showcasing several proof-of-concept projects.

Chinar’s notable project included in this edition,

”The Swachha Shala Initiative,” focuses on improv-

ing hygiene in schools across India, particularly for

girls whose attendance is aﬀected by inadequate facili-

ties. The initiative employs a React Native app that en-

ables students to upload images of school bathrooms,

which are then analyzed by a Convolutional Neural

Network (CNN) to detect cleanliness issues. By in-

volving students in maintaining hygiene standards, the

project aims to create a healthier learning environment

and reduce school absences. Currently, the initiative

is in the testing phase, with ongoing eﬀorts to enhance

its machine-learning models.

In the article ”Transformer architecture in Com-

puter vision-Part II” by Blesson George, he continues

the discussion about Vision transformers. He discuss

about the seminal paper ”An Image is Worth 16x16

Words: Transformers for Image Recognition at Scale”

by Alexey Dosovitskiy et al. This discussion helps

to understand how the transformer architecture revo-

lutionized the image recognition tasks and understand

the innovative techniques introduced in this paper.

The article of Aromal P, ”X-ray Astronomy: Through

Missions” explores the development and achievements

of X-ray astronomy through various satellite missions.

Early X-ray satellites could not directly image X-ray

sources due to the challenges of focusing high-energy

photons. The launch of the Einstein satellite in 1978

marked a breakthrough with its innovative focusing

optics, dramatically enhancing the sensitivity and res-

olution of X-ray imaging. Subsequent missions in the

1980s, including Japan’s Hinotori and Tenma, the Eu-

ropean Space Agency’s EXOSAT, and the Japanese-

British-American collaboration Ginga, further advanced

the ﬁeld with discoveries in solar phenomena, high-

energy X-ray sources, and cosmic phenomena like

quasars and X-ray binaries. These missions set the

stage for a ”golden period” of X-ray astronomy from

the 1990s onwards, characterized by even more sensi-

tive and capable satellites. Future articles will delve

into these groundbreaking missions.

Sindhu G examines in the article ”Exploring Stel-

lar Clusters: Insights from Color-Magnitude Diagrams,

Part-2” the Hertzsprung-Russell diagram (HRD) of the

globular star cluster Messier 55. It highlights how HR

diagrams oﬀer insights into stellar evolution by com-

paring star clusters with diﬀerent chemical composi-

tions and evolutionary stages. The HRD of Messier 55

shows it is an older cluster with a higher proportion of

evolved stars and a lack of high-mass main-sequence

stars, such as types O, B, A, and F. The presence of

blue stragglers and red giants indicates stellar mergers

and diﬀerent evolutionary phases. The article contrasts

these features with those of younger open clusters and

suggests further exploration of HR diagrams of various

clusters in future articles.

Geetha Paul explores. in the article ”Drug Re-

purposing or Drug Repositioning: An Emerging Ap-

proach in Drug Discovery”, the concept of drug re-

purposing, is a strategy to ﬁnd new therapeutic uses

for existing drugs. Unlike the traditional drug discov-

ery process, which is time-consuming and costly, drug

repurposing leverages known safety proﬁles and clini-

cal data to reduce development time, costs, and risks.

The article discusses various methodologies for drug

repurposing, including drug-oriented, target-oriented,

and disease/therapy-oriented approaches. This strat-

egy has proven eﬀective for treating rare, complex, and

neglected diseases and has gained prominence due to

advancements in bioinformatics, cheminformatics, and

artiﬁcial intelligence. The article concludes that drug

repurposing is becoming an essential component of

modern drug discovery.

The article ”The Evolution of Artiﬁcial Intelli-

gence in Neonatology: An India-Focused Perspec-

tive” by Kalyani Bagri explores the growing role of

AI in improving neonatal care. Globally, AI is trans-

forming Neonatal Intensive Care Units (NICUs) by

enhancing monitoring, diagnosis, and treatment. In

India, while the adoption has been slower, innova-

tive and cost-eﬀective AI solutions tailored to local

challenges are gaining traction. AI technologies such

as predictive analytics, computer vision, and Natural

Language Processing (NLP) are being used to pre-

dict complications, enable timely interventions, and

provide real-time data-driven insights. Despite chal-

lenges like fragmented infrastructure and data quality

issues, initiatives by prominent Indian organizations

and collaborations among government and private en-

tities are driving AI integration in neonatology. These

eﬀorts aim to improve patient outcomes and expand

access to specialized care, particularly in underserved

regions, by enhancing operational eﬃciency in elec-

tronic NICUs (e-NICUs). The article ”Bead Chip

Technology: Simplifying Genetic Analysis” by Jinsu

Ann Mathew explains how DNA microarray technol-

ogy has revolutionised genetic analysis by allowing the

simultaneous study of thousands of genes. This tech-

nology uses a chip with thousands of speciﬁc DNA

sequences (probes) to detect and analyze genetic varia-

tions and gene expressions in a sample. Diﬀerent types

of DNA microarrays—spotted arrays, self-assembled

arrays, and in-situ synthesized arrays—oﬀer unique ad-

vantages for various genetic studies, from gene expres-

sion proﬁling to identifying genetic mutations. Each

method uses innovative approaches like hybridization,

photolithography, and inkjet printing to enhance preci-

sion and eﬃciency in genetic analysis. As microarray

technology advances, it holds the potential to deepen

our understanding of genetics and support the develop-

ment of personalized medicine.

The article ”The Evolution of GPUs: Powering

the Next Generation of High-End Computing and Ar-

tiﬁcial Intelligence” by Atharva Pathak explores the

transformation of GPUs from specialized graphics hard-

ware to essential tools in modern computing. Ini-

tially developed to handle complex graphics rendering,

GPUs have evolved into versatile processors capable of

massively parallel computations. This shift has made

them crucial for high-performance tasks like scientiﬁc

research, AI, and machine learning. Key innovations,

such as NVIDIA’s CUDA and Tensor Cores, have en-

hanced their capabilities in AI training. Despite chal-

lenges like high power consumption and programming

complexity, GPUs are poised to drive future advance-

ments in AI and computing, with ongoing development

toward more powerful and eﬃcient architectures.

iii

News in Brief

by News Desk

airis4D, Vol.2, No.9, 2024

www.airis4d.com

Dr Kaveri Kale was awarded Phd for here thesis titled Automatic Radiology Report Generation for

Diﬀerent Radiology Modalities by Institute: Indian Institute of Information Technology, Bombay. She worked

under the supervision of Prof. Pushpak Bhattacharyya, Department of Computer Science and Engineering, IIT

Bombay and external supervisor Prof. Ninan Sajeeth Philip, airis4D.

News in Brief

by News Desk

airis4D, Vol.2, No.9, 2024

www.airis4d.com

Geetha Paul and Jinsu Ann Mathews with students after an outreach activity for students at SNV UP

School, Thadiyoor.

Contents

Editorial ii

News in Brief v

News in Brief vi

I Artiﬁcial Intelligence and Machine Learning 1

1 Transformer Architecture in Computer Vision-Part II 2

1.1 Vision Transformer (ViT) Concept . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2

1.2 Transformer Encoder in the Vision Transformer (ViT) . . . . . . . . . . . . . . . . . . . . . . . . 3

1.3 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4

II Astronomy and Astrophysics 5

1 When the Crab Slows Down : A Bharathiya Antariksh Hackthon Story 6

1.1 Bharathiya Antariksh Hackathon 2024 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

1.2 The Crab is Slowing Down . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

2 The Pleiades Cluster 10

2.1 Pleiades . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

2.2 Mythological Signiﬁcance of the Pleiades . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

2.3 Formation of the Pleiades Cluster . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

2.4 Structure of the Pleiades Cluster . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

2.5 Locating the Pleiades . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

2.6 Astronomical Signiﬁcance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

2.7 Astrophysical Importance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

2.8 Cultural and Historical Signiﬁcance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

2.9 Visibility and Aesthetic Appeal . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

2.10 Scientiﬁc Discoveries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

III Biosciences 14

1 Drug Repurposing or Drug Repositioning: An Emerging Approach in Drug Discovery 15

1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

1.2 Traditional Drug Discovery vs. Drug Repurposing . . . . . . . . . . . . . . . . . . . . . . . . . . 16

1.3 Advantages of Drug Repurposing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16

1.4 Methodologies of Drug Repurposing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17

2 The Evolution of Artiﬁcial Intelligence in Neonatology: An India-Focused Perspective 20

2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20

CONTENTS

2.2 AI and Neonatology: A Growing Synergy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20

2.3 AI in Electronic Neonatal Intensive Care Units (e-NICU) . . . . . . . . . . . . . . . . . . . . . . 21

3 Biomarkers: Powerful Indicators in Healthcare 23

3.1 Diagnostic Biomarkers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23

3.2 Prognostic Biomarkers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

3.3 Predictive Biomarkers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

3.4 Risk Biomarkers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

3.5 Safety Biomarkers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

3.6 Monitoring Biomarkers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

3.7 Summary of Biomarker Types . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26

3.8 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26

IV General 28

1 The Swachha Shala Initiative 29

1.1 The Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29

1.2 My Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29

1.3 Dataset . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30

1.4 Current App . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30

1.5 Epilogue . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30

V Computer Programming 32

1 The Evolution of GPUs: Powering the Next Generation of High-End Computing and

Artiﬁcial Intelligence 33

1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33

1.2 The Birth of the GPU: From Graphics to General-Purpose Processing . . . . . . . . . . . . . . . 33

1.3 GPU Architecture: A Deep Dive . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33

1.4 From Graphics to General-Purpose Computing . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34

1.5 GPUs in AI and Machine Learning: A Match Made in Silicon . . . . . . . . . . . . . . . . . . . 34

1.6 Challenges and Limitations of GPUs in High-End Computing . . . . . . . . . . . . . . . . . . . 34

1.7 The Future of GPUs: What’s Next? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35

1.8 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35

1.9 Suggested Further Reading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35

1.10 Timeline of GPU Evolution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35

1.11 GPU Architecture Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36

1.12 GPUs in Action . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36

1.13 Future of GPUs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36

viii

Part I

Artiﬁcial Intelligence and Machine Learning

Transformer Architecture in Computer

Vision-Part II

by Blesson George

airis4D, Vol.2, No.9, 2024

www.airis4d.com

In the previous issue, we explored the core prin-

ciples of Vision Transformers (ViTs) and their adapta-

tion of the traditional transformer architecture to han-

dle image data. In this issue, we will take a closer

look at the seminal paper ”An Image is Worth 16x16

Words: Transformers for Image Recognition at Scale”

by Alexey Dosovitskiy et al. This discussion will in-

volve a deep dive into the paper to understand how

it revolutionizes image recognition by applying trans-

formers, originally designed for text, to visual data.

We’ll explore the innovative techniques introduced in

this paper, including the division of images into patches

and the use of self-attention to capture complex pat-

terns, and discuss the broader implications of this ap-

proach for the ﬁeld of computer vision.

In recent years, traditional convolutional neural

networks (CNNs) have dominated the ﬁeld of image

recognition; however, they come with inherent limi-

tations that can hinder their scalability and eﬃciency.

Despite their success, as demonstrated by foundational

works such as LeCun et al. (1989) and Krizhevsky

et al. (2012), CNNs often rely on inductive biases

that may not be optimal for all tasks. As the demand

for more sophisticated models grows, it becomes in-

creasingly important to explore alternative architec-

tures that can better handle the complexities of vi-

sual data. This paper introduces the Vision Trans-

former (ViT), which leverages the Transformer archi-

tecture—originally designed for natural language pro-

cessing (NLP) by Vaswani et al. (2017)—to address

these challenges in computer vision. By applying the

principles of self-attention and treating images as se-

quences of patches, ViT interprets an image as a se-

quence of ﬂattened 2D patches, allowing it to process

visual information in a manner similar to how Trans-

formers handle text. This novel approach not only en-

hances performance on image classiﬁcation tasks but

also demonstrates that a pure Transformer can achieve

state-of-the-art results, thereby opening new avenues

for research in the application of Transformers beyond

text-based data.

1.1 Vision Transformer (ViT)

Concept

The Vision Transformer (ViT) concept represents

a signiﬁcant shift in how images are processed for tasks

such as image classiﬁcation. Unlike traditional convo-

lutional neural networks (CNNs), which rely on con-

volutional layers and local receptive ﬁelds to extract

features from images, ViT employs the Transformer

architecture, originally designed for natural language

processing (NLP). Here are the key components of the

ViT concept:

Image as a Sequence of Patches: ViT begins by

dividing an input image into ﬁxed-size patches.

For example, an image might be split into 16x16

pixel patches. Each patch is then ﬂattened into

a one-dimensional vector, eﬀectively transform-

ing the 2D image into a sequence of 1D patch

embeddings. This approach allows the model to

1.2 Transformer Encoder in the Vision Transformer (ViT)

treat the image similarly to how text is processed

in NLP, where words are treated as tokens in a

sequence.

Patch Embeddings: Each ﬂattened patch is pro-

jected into a higher-dimensional space using a

trainable linear projection. This process gener-

ates what are known as patch embeddings, which

serve as the input tokens for the Transformer

model. The dimensionality of these embeddings

is consistent throughout the model, allowing for

eﬃcient processing.

Self-Attention Mechanism: The core of the

Transformer architecture is the self-attention mech-

anism, which enables the model to weigh the

importance of diﬀerent patches relative to one

another. This allows ViT to capture long-range

dependencies and relationships between diﬀer-

ent parts of the image, which is crucial for un-

derstanding complex visual patterns.

Transformer Encoder: ViT utilizes a standard

Transformer encoder, which consists of multi-

ple layers of self-attention and feed-forward net-

works. Each layer processes the sequence of

patch embeddings, allowing the model to learn

hierarchical representations of the image. The

architecture is designed to be scalable, making

it possible to train large models on extensive

datasets.

Pre-training and Fine-tuning: Similar to NLP

applications, ViT beneﬁts from a two-step train-

ing process: pre-training on a large dataset (such

as ImageNet-21k or JFT-300M) followed by ﬁne-

tuning on a smaller, task-speciﬁc dataset. This

approach allows the model to learn general fea-

tures during pre-training, which can then be adapted

to speciﬁc tasks during ﬁne-tuning.

1.2 Transformer Encoder in the

Vision Transformer (ViT)

The Transformer encoder is a fundamental part of

the Vision Transformer architecture, designed to pro-

cess the sequence of patch embeddings derived from an

Figure 1: Figure 1 provides a comprehensive overview

of the Vision Transformer architecture. The process

begins with the input image, which is divided into

ﬁxed-size patches. Each of these patches is then ﬂat-

tened and linearly projected into a lower-dimensional

space to create patch embeddings. To incorporate

spatial information, position embeddings are added to

these patch representations. The resulting sequence of

embedded patches, along with an additional learnable

classiﬁcation token, is fed into a standard Transformer

encoder. This encoder consists of multiple layers that

utilize multi-head self-attention and feed-forward net-

works to process the sequence, ultimately enabling the

model to learn rich representations of the image for

classiﬁcation tasks.

input image. Each encoder layer plays a crucial role in

capturing the relationships and dependencies between

diﬀerent patches, allowing the model to learn rich and

meaningful representations of the image.

Within each layer, the ﬁrst component is the multi-

head self-attention mechanism. This mechanism en-

ables the model to focus on various parts of the in-

put sequence simultaneously. By computing attention

scores for each patch in relation to all other patches,

the model can determine the importance of each patch

based on its context. The multi-head aspect allows

multiple attention mechanisms to operate in parallel,

capturing diﬀerent types of relationships and features.

Following the self-attention mechanism, the out-

put is processed by a feed-forward neural network. This

network consists of two linear transformations with a

non-linear activation function in between, further reﬁn-

ing the representations learned from the self-attention

layer. To enhance training stability and convergence,

each sub-layer is equipped with a residual connection

and layer normalization. The residual connection adds

the input of the sub-layer to its output, which helps

mitigate issues like vanishing gradients, while layer

1.3 Conclusion

normalization standardizes the outputs for stable train-

ing dynamics.

The ViT model stacks several of these Trans-

former encoder layers, allowing each layer to build

upon the output of the previous one. This stacking en-

ables the model to learn increasingly abstract and com-

plex features of the input image. To maintain spatial

awareness, positional encodings are added to the patch

embeddings, providing information about the relative

positions of the patches in the original image.

The ﬁnal output of the Transformer encoder is a

set of embeddings that encapsulate the learned rela-

tionships and features of the entire image. This output

can then be utilized for various tasks, such as classiﬁ-

cation, by applying a classiﬁcation head to the output

corresponding to a speciﬁc patch, often referred to as

the ”class token.”

1.3 Conclusion

Vision Transformer (ViT) represents a signiﬁcant

advancement in the ﬁeld of computer vision by reimag-

ining how images are processed for tasks like classi-

ﬁcation. By leveraging the Transformer architecture,

originally designed for natural language processing,

ViT demonstrates that self-attention mechanisms can

eﬀectively capture the intricate relationships within vi-

sual data. The novel approach of treating images as

sequences of patches, combined with the power of the

Transformer encoder, enables ViT to achieve state-of-

the-art performance without relying on the traditional

convolutional layers of CNNs. This paradigm shift not

only challenges the dominance of CNNs in computer

vision but also opens up new possibilities for apply-

ing transformers to a wide range of visual tasks. As

research in this area continues to evolve, the Vision

Transformer is poised to inspire further innovations

and improvements in both the theory and application

of deep learning models for image recognition and be-

yond.

References

1. Dosovitskiy, Alexey, et al. ”An image is worth

16x16 words: Transformers for image recogni-

tion at scale.” arXiv preprint arXiv:2010.11929

(2020).

2. Vision Transformers

About the Author

Dr. Blesson George presently serves as an

Assistant Professor of Physics at CMS College Kot-

tayam, Kerala. His research pursuits encompass the

development of machine learning algorithms, along

with the utilization of machine learning techniques

across diverse domains.

Part II

Astronomy and Astrophysics

When the Crab Slows Down : A Bharathiya

Antariksh Hackthon Story

by Aromal P

airis4D, Vol.2, No.9, 2024

www.airis4d.com

1.1 Bharathiya Antariksh Hackathon

2024

Honorable Prime Minister of India Shri Narendra

Modi declared 23

August as the National Space Day,

to celebrate the success in achieving a soft landing

of the Vikram Lander at the Shiv Shakti point on the

southern pole of the moon on 23

August 2023. The

nation celebrated its ﬁrst National Space Day (NSpD)

on 23

August 2024. A series of events were held

across the nation as a part of the celebration and two

national level competitions were conducted by ISRO:

Bharathiya Antariksh Hackathon 2024 (BAH -

2024).

ISRO Robotic Challenge.

Here, I would like to emphasise more on BAH - 2024.

This nationwide competition was launched by ISRO in

collaboration with Hack2skill on 4

July 2024. The

competition. 12 problem statements were given:

Optimizing Urban Futures: Leveraging Digital

Twins for Comprehensive Infrastructure Man-

agement.

Generation of Rooftop Solar Energy Potential

Map Using Machine Learning/Deep Learning

Based Building Footprint Extraction.

Automatic detection of craters & boulders from

Orbiter High Resolution Camera(OHRC) images

using AI/ML techniques.

Voice enabled user interface for geospatial map

based.

Nowcasting of Precipitation Systems using C-

Band Doppler Radar Observations.

Feature Extraction from Remote Sensing High

Resolution Data using AI/ML (Ex–High Tension

tower, windmill, electric substation, Brick Kiln,

farmbunds).

How much the Crab pulsar has slowed down

since launch of AstroSat? Finding rate of spin

down rate of Crab pulsar with AstroSat LAXPC

and CZTI observations.

Spectral classiﬁcation of Chandrayaan-2 IIRS

using AI/ML for understanding geological di-

versity of the Moon.

Identiﬁcation of safe navigation routes on the

Moon using Chandrayaan Images.

Image based Search of Lunar craters from global

mosaic.

Lunar surface image simulation and Visualiza-

tion.

Context-Aware Geospatial Data Retrieval using

LLM/NLP.

A team of 3-4 members can be formed to propose an

idea and solve it. After the submission of the ideas,

around 100 teams were selected for the initial inter-

view, which was held online and from that 30 teams

were selected for the Grand Finale, which was held on

13-14 August 2024 at the NRSC oﬃce, Hyderabad. At

the end of this 30 hour hackathon, 12 teams were se-

lected to present their work from which 3 temas were

declared as the winners and 2 teams were given con-

solation prizes. The top three teams were invited to

1.2 The Crab is Slowing Down

Figure 1: Team Khagola at Bharath Mandapam in

Delhi. From Left, Himanshu Grover, Shyam Praksh,

Aromal P and Biki Ram.

the maiden National Space Day celebration held at

Bharath Mandapam in New Delhi, where they pre-

sented their results and received the award from the

Honorable President of India Smt. Droupadi Murmu.

In this article, I would like to emphasise the sci-

ence that led team Khagola to win the third posi-

tion in Bharathiya Antariksh Hackthon-2024. Team

Khagola(Figure 1), formed by 4 PhD students - Hi-

manshu Grover (IIT Roorkee), Shyam Prakash (URSC

Banglore), Biki Ram (IIT Indore), and myself, Aromal

P (IIT Indore), worked on the problem statement ”How

much the Crab pulsar has slowed down since launch of

AstroSat? Finding rate of spin down rate of Crab pul-

sar with AstroSat LAXPC and CZTI observations?”

1.2 The Crab is Slowing Down

Neutron stars are the stellar remnants of interme-

diate mass stars which were too massive to form white

dwarfs but not massive enough to form a black hole.

They serve as a natural laboratory to study high-energy

and extremely dense phenomena. Formed from stars

whose radius spans over millions of kilometers, these

stars are usually found to have a radii 10km and a

mass of about 1.4 times that of the Sun. When they

are formed, they rotate at very high speed to conserve

the angular momentum. Over time, the rotating neu-

tron star, also known as the pulsar, gradually loses its

energy and begins to slow down. However, the decel-

eration is so gradual that it takes thousands of years

for their rotational period to slow down by 1 second.

As a result, the pulsars are regarded as one of the most

stable rotators in the universe.

Crab, an isolated neutron star situated at the heart

of the Crab Nebula, is believed to have been formed

in 1054 AD after a supernova explosion, which was

recorded for the ﬁrst time in the human history. Crab

is one of the brightest objects in the X-ray wavelength

and is often used as a standard reference for brightness

in high energy astrophysics. Crab has been studied for

decades in both radio and X-ray wavelengths. Most

high energy experiments rely on the Crab Pulsar for

time calibrations.

1.2.1 Crab Pulsar Data of AstroSat

In order to ﬁnd the spin down rate and the rate of

spin down rate of a pulsar, it is necessary to have an

instrument with high timing capabilities. AstroSat, In-

dia’s ﬁrst multi-wavelength satellite, is one such satel-

lite which has two payloads - a Large Area X-ray Pro-

portional Counters (LAXPC) and a Cadmium-Zinc-

Telluride Imager (CZTI) with high timing capabilities.

LAXPC has an eﬀective area of nearly 10,000 cm

and

works on 3-80 keV energy range. CZTI on the other

hand, has an eﬀective area of 1,000 cm

and works on

20-200 keV energy range. From its launch in 2015, the

AstroSat has been observing Crab pulsar and collecting

data regularly for calibration purpose.

The AstroSat LAXPC and CZTI data can be re-

duced with the help of LaxpcSoft Format-B software

and New CZTI Pipeline version 3, respectively. For the

precise timing analysis, all timing information should

be modiﬁed with respect to the center of the solar sys-

tem. For that, the barycentric correction is also re-

quired, which can be done with the help of the as1bary

codes. For around 900 orbits of data to be reduced man-

ually, it will take a lot of time. Our team automated all

these steps using a script during this hackathon, pro-

viding a better and eﬃcient way to reduce these data.

1.2.2 Leahy Power Spectrum and Z

Statistics Finds the Period!

We used the Leahy power spectrum, which uses

Fast Fourier Transform (FFT) Algorithm to ﬁnd the

1.2 The Crab is Slowing Down

diﬀerent frequencies in a signal. FFT is used to com-

pute the discrete Fourier transform. To ﬁnd out the

oscillations using FFT, binning of data is required and

it will not considering the harmonics. Z

statistics is

another approach that we used to ﬁnd the oscillations

in a signal. This is similar to that of Fourier transform

but this does not require binning of the data and also

considers the eﬀect of harmonics.

AstroSat completes one revolution around the earth

in 90 minutes. This provides us a window of 45 min-

utes to continuously observe an object. In one orbit,

we will get around 3,000 seconds of exposure. Using

the Leahy power spectrum, we will be getting less pre-

cise value for frequency. For more precise values, Z

method is a better approach but they are computation-

ally expensive for a long range of frequencies. So, we

used both FFT and Z

statistics, to ﬁnd a more pre-

cise value of the rotational frequency. Combining both

methods, we have reached upto 7 signiﬁcant ﬁgures.

1.2.3 Let’s ﬁnd the spin down rate

We calculated the frequency and period of se-

lected orbits from each year, starting from 2015 to 2022

(publicly available data) using both the instruments -

LAXPC and CZTI, and plotted it against time. We then

compared our results with the already available radio

data from Jodrell Bank Center for astrophysics (JBO).

From the plot, ﬁtting it with Taylor expansion, we cal-

culated the spin down rate and rate of spin down rate,

as shown in ﬁgure(2). These values that we calculated

matched exactly with the already available radio data.

1.2.4 We’ve got a Glitch......!

We also tried to plot the rate of change of fre-

quency over time (Figure 3). Then, we came across

an event called Glitch, the abrupt changes in rotation

rate of pulsar for a short period of time. After Glitch,

the pulsar will recover and follow its previous state.

It was the ﬁrst ever discovery of a Glitch event from

the AstroSat data. It is suggested that the Glitches are

arising from the internal properties of a neutron star

and hence it can be used to constrain the equation of

Figure 2: Period vs time and frequency vs time graph

of crab pulsar from LAXPC and CZTI data compared

against Jodrell Bank Observatory data

1.2 The Crab is Slowing Down

Figure 3: Change in frequqncy vs time graph of crab

pulsar from LAXPC and CZTI data compared against

Jodrell Bank Observatory data, vertical lines represents

the glitch event observed from radio measurements

states of neutron stars.

All these ﬁndings are done within 30 hr BAH

ﬁnale held at National Remote Sensing Center, Hyder-

abad on August 13-14 2024. We are carrying out more

studies and will be updated.

About the Author

Aromal P is a research scholar in Depart-

ment of Astronomy, Astrophysics and Space Engi-

neering (DAASE) in Indian Institute of Technology

Indore. His research mainly focuses on studies of

Thermonuclear X-ray Bursts on Neutron star surface

and its interaction with the Accretion disk and Corona.

The Pleiades Cluster

by Sindhu G

airis4D, Vol.2, No.9, 2024

www.airis4d.com

2.1 Pleiades

The Pleiades cluster (Figure: 1 and Figure: 2),

also known as the Seven Sisters or Messier 45, is an

open star cluster located in the constellation Taurus. It

is one of the closest and most visible star clusters to the

naked eye, even from areas with moderate light pollu-

tion. The Pleiades is visible from nearly every location

on Earth, from the North Pole to areas well beyond

the southernmost tip of South America. The clus-

ter is approximately 440 light-years away from Earth

and consists of several hundred stars, although only a

few of the brightest ones are easily visible without a

telescope. The cluster’s total mass is estimated to be

around 800 solar masses, with a core radius of roughly

8 light-years and a tidal radius reaching approximately

43 light-years. The Pleiades is distinguished by its

vivid blue color, produced by the high temperatures of

its prominent B-type stars, which are relatively young,

with an estimated age of 75 to 150 million years. The

Pleiades is best viewed in the Northern Hemisphere

during the winter months, especially from October to

April.

2.2 Mythological Signiﬁcance of the

Pleiades

In Greek mythology, the Pleiades are the seven

daughters of the Titan Atlas and the ocean nymph

Pleione. Their names—Alcyone, Celaeno, Electra,

Maia, Merope, Sterope, and Taygete—are deeply em-

bedded in myth. According to legend, Zeus trans-

formed these nymphs into stars to protect them from

Figure 1: The Pleiades is an open star cluster in the

constellation Taurus. (Image Credit: Manfred Konrad

via Getty Images)

Figure 2: Pleiades, identikit of the brightest star clus-

ter. (Image Credit: Wikipedia)

2.3 Formation of the Pleiades Cluster

the hunter Orion. This transformation led to the clus-

ter being known as the ”Seven Sisters.” The Pleiades

have been signiﬁcant across various cultures, serving

as markers for seasonal changes and guiding ancient

mariners in their navigation. While the cluster is most

commonly associated with these seven prominent stars,

additional stars within the Pleiades can be seen with the

aid of binoculars or telescopes.

2.3 Formation of the Pleiades Cluster

Origin: The Pleiades cluster formed about 100

million years ago from a large, dense cloud of gas and

dust in the interstellar medium. This cloud underwent

a process of gravitational collapse, leading to the for-

mation of numerous stars in a relatively short period.

Triggering Mechanism: The initial collapse of

the cloud may have been triggered by nearby super-

novae or the inﬂuence of other massive stars, which

caused shock waves that compressed the gas and dust,

initiating star formation.

Stellar Birth: The stars in the Pleiades formed

from the remaining gas and dust in the cluster’s core.

The cluster’s young age means that most of its stars are

still on the main sequence of stellar evolution, where

they are actively fusing hydrogen into helium.

2.4 Structure of the Pleiades Cluster

2.4.1 Core and Halo:

Core: The core of the Pleiades cluster is densely

packed with young, massive stars. This central region

is where most of the cluster’s bright, blue stars are

found. The core radius is approximately 8 light-years.

Halo: Surrounding the core is a more diﬀuse halo of

stars. The tidal radius, which deﬁnes the region within

which the cluster’s gravitational inﬂuence is dominant,

extends to about 43 light-years.

2.4.2 Star Distribution:

Main Sequence: The majority of the stars in

the Pleiades are on the main sequence. These stars are

mostly of spectral types B and A, characterized by their

blue and white colors due to their high temperatures.

Pre-Main-Sequence Stars: Some stars in the Pleiades

are still in their pre-main-sequence phase, where they

are contracting and heating up before beginning hydro-

gen fusion.

2.4.3 Nebula:

Reﬂection Nebula: The Pleiades is surrounded

by a faint reﬂection nebula, which is a cloud of in-

terstellar dust that reﬂects the light from the cluster’s

bright stars. This nebula is visible in long-exposure

photographs and contributes to the cluster’s character-

istic blue appearance.

2.4.4 Stellar Dynamics:

Gravitational Interactions: The stars in the Pleiades

interact gravitationally, but the cluster is not densely

packed like some older clusters. This relatively loose

conﬁguration is typical for a young open cluster.

Evolution: As the Pleiades ages, its stars will gradu-

ally drift apart due to tidal interactions with the Milky

Way and the internal dynamics of the cluster. Over

time, the cluster will disperse and become less densely

populated.

2.4.5 Distance and Size:

Distance: The Pleiades is located approximately

440 light-years from Earth, making it one of the closest

star clusters to our solar system.

Size: The cluster spans about 20 light-years across,

with its stars spread out in a relatively compact and

well-deﬁned region.

2.5 Locating the Pleiades

To locate the Pleiades(Figure 3), begin by ﬁnding

the well-known constellation Orion, the hunter. Use

the three stars in Orion’s belt to draw a line upward,

extending beyond his bow. The ﬁrst bright star you’ll

encounter is Aldebaran, which represents the eye of the

bull in the constellation Taurus. According to Earth-

Sky, Aldebaran means ”follower” in Arabic, as it fol-

lows the Pleiades across the sky. The Pleiades cluster

2.6 Astronomical Signiﬁcance

Figure 3: The Pleiades are easy to locate. (Image

Credit: Daisy Dobrijevic)

will be just beyond Aldebaran, appearing as a small

dipper-shaped group of stars.

2.6 Astronomical Signiﬁcance

Stellar Evolution Studies: The Pleiades is a rel-

atively young open star cluster, estimated to be around

100 million years old. This makes it an ideal laboratory

for studying the early stages of stellar evolution. By

examining its stars, astronomers can gain insights into

how stars form, evolve, and interact within a cluster.

Distance Calibration: The Pleiades has been cru-

cial for calibrating the distance scale of the universe.

Accurate distance measurements to the Pleiades help

reﬁne techniques used to measure distances to other

star clusters and galaxies, which is fundamental for

understanding the size and structure of the universe.

2.7 Astrophysical Importance

Understanding Star Formation: As a young

cluster, the Pleiades contains many hot, blue stars.

Studying these stars helps astronomers understand the

processes involved in star formation and the life cycles

of stars. The presence of a reﬂection nebula around the

cluster also provides information about the interaction

between starlight and interstellar dust.

Stellar Dynamics: The cluster’s relatively close prox-

imity allows for detailed studies of stellar motions and

dynamics within a cluster. Observing how stars move

within the Pleiades helps scientists understand the grav-

itational interactions and evolution of star clusters.

2.8 Cultural and Historical

Signiﬁcance

Mythology and Folklore: The Pleiades has been

a signiﬁcant feature in the mythology and folklore of

various cultures worldwide, from the Greek myth of

the Seven Sisters to Native American stories and be-

yond. These myths have enriched cultural narratives

and contributed to the understanding of how ancient

peoples perceived the sky.

Seasonal Marker: Historically, the Pleiades has been

used as a marker for agricultural and navigational pur-

poses. Its rising and setting were often associated with

the timing of planting and harvesting crops, as well as

guiding ancient sailors.

2.9 Visibility and Aesthetic Appeal

Easily Visible: The Pleiades is one of the most

easily recognizable star clusters, visible to the naked

eye even in moderately light-polluted skies. Its bright

stars and compact arrangement make it a popular object

for stargazers and amateur astronomers.

Aesthetic Beauty: The cluster’s striking blue stars

and the surrounding reﬂection nebula create a visually

stunning spectacle in the night sky, making it a favorite

target for astrophotographers.

2.10 Scientiﬁc Discoveries

Exoplanet Studies: The Pleiades cluster has also

been the subject of studies looking for exoplanets. Un-

derstanding the types of planets that might form around

young stars can provide insights into planetary forma-

tion and the potential for life elsewhere in the universe.

Overall, the Pleiades cluster’s proximity, visibil-

ity, and rich stellar population make it a key object of

study in both professional and amateur astronomy, as

well as a signiﬁcant part of human cultural heritage.

By examining the Hertzsprung-Russell diagram

and color-magnitude diagrams of the Pleiades, astronomers

can study how stars of diﬀerent masses evolve over time

and gain insights into the processes that govern star for-

mation and the early stages of stellar life.

2.10 Scientiﬁc Discoveries

In the upcoming article, we will delve into the

Hertzsprung-Russell diagram of the Pleiades cluster.

References:

Star Clusters

Pleiades

The Pleiades – or 7 Sisters – known around the

world

HR Diagram, Star Clusters,and Stellar Evolution

The Pleiades: Facts about the ”Seven Sisters”

star cluster

Pleiades, identikit of the brightest star cluster

Pleiades

About the Author

Sindhu G is a research scholar in Physics

doing research in Astronomy & Astrophysics. Her

research mainly focuses on classiﬁcation of variable

stars using diﬀerent machine learning algorithms. She

is also doing the period prediction of diﬀerent types

of variable stars, especially eclipsing binaries and on

the study of optical counterparts of X-ray binaries.

Part III

Biosciences

Drug Repurposing or Drug Repositioning:

An Emerging Approach in Drug Discovery

by Geetha Paul

airis4D, Vol.2, No.9, 2024

www.airis4d.com

(Image courtesy:

https://www.slideshare.net/slideshow/drug-repurposing-147759373/147759373)

1.1 Introduction

Drug repurposing is a strategy aimed at discov-

ering new therapeutic indications for existing drugs,

including those already on the market, used in various

clinical settings, or highly characterised compounds—

even those that were previously considered failures.

Recently, drug repurposing has gained prominence as

an alternative approach for rapidly identifying and de-

veloping new pharmaceuticals, particularly for rare and

complex diseases that currently lack eﬀective treat-

ments. This review will explore the current status

of drug repurposing for various diseases, including

skin disorders, infectious diseases, inﬂammatory con-

ditions, cancers, and neurodegenerative diseases. Ad-

ditionally, the review will provide insights into these

repurposed drugs’ structural features and mechanisms

of action.

Traditional drug discovery is a well-established

process that identiﬁes and develops New Molecular

Entities (NMEs) to treat various diseases. This ap-

proach typically involves several sequential stages: dis-

covery and preclinical research, safety review, clinical

research, regulatory review (such as FDA review), and

post-market safety monitoring. The process begins

with identifying a potential therapeutic target, followed

by the design and synthesis of compounds that can

interact with this target. These compounds undergo

extensive preclinical testing, including in vitro and in

vivo studies, to assess their eﬃcacy, safety, and phar-

macokinetic properties. Once a promising compound

is identiﬁed, it proceeds to clinical trials, divided into

several phases (Phase I, II, and III) to evaluate its safety

and eﬃcacy in humans. If the compound successfully

passes these clinical trials, it is submitted for regulatory

review. The drug enters the market upon approval, but

continuous monitoring is required to ensure its long-

term safety and eﬃcacy.

The traditional drug discovery process is notori-

ously lengthy, time-consuming, and expensive. It can

take anywhere from 10 to 16 years for a new drug to

reach the market, with an estimated cost of around $12

billion. Moreover, the risk of failure is high, with many

compounds failing at various stages of development

due to issues such as lack of eﬃcacy, unacceptable

toxicity, or poor pharmacokinetic properties.

Drug repurposing, also known as drug reposition-

ing, is an alternative approach to drug discovery that

focuses on ﬁnding new therapeutic uses for existing

drugs. This strategy leverages the vast amount of data

and knowledge accumulated during the development

and clinical use of approved drugs to identify new in-

dications for these compounds. Drug repurposing typ-

1.2 Traditional Drug Discovery vs. Drug Repurposing

ically involves four main stages: compound identiﬁ-

cation, compound acquisition, development, and post-

market safety monitoring. Unlike traditional drug dis-

covery, drug repurposing does not require the initial

stages of target identiﬁcation and compound design,

as it starts with existing drugs that have already been

proven safe for human use. This signiﬁcantly reduces

the time and cost associated with drug development.

Drug repurposing can be approached through vari-

ous methodologies, including drug-oriented, target-

oriented, and disease/therapy-oriented strategies. Drug-

oriented methods focus on the structural characteristics

and biological activities of drugs, while target-oriented

methods involve screening drugs against speciﬁc bio-

logical targets. Disease/therapy-oriented methods utilise

information about the disease process, such as pro-

teomics, genomics, and metabolomics data, to identify

potential new uses for existing drugs.

(Image courtesy:

https://pubs.acs.org/journal/acsodf?ref=pdf )

Figure 1: Illustration of the steps involved in the drug

repurposing process. Step 1: Screening − identifying

potential repurposed drug candidates from a large pool

of drugs using appropriate computational or experi-

mental methodologies, Step 2: Shortlisting − selection

of potential lead compounds, Step 3: Validation −

validating the discovered drug through preclinical and

clinical trial investigations.

1.2 Traditional Drug Discovery vs.

Drug Repurposing

The traditional drug discovery process is lengthy,

time-consuming, and expensive, involving ﬁve stages:

discovery and preclinical, safety review, clinical re-

search, FDA review, and FDA post-market safety mon-

itoring. This process can take 10-16 years and cost

around $12 billion, with a high risk of failure. In

contrast, drug repurposing involves only four stages:

compound identiﬁcation, compound acquisition, de-

velopment, and FDA post-market safety monitoring.

With advancements in bioinformatics and cheminfor-

matics tools and the availability of extensive biological

and structural databases, drug repurposing has signiﬁ-

cantly reduced the time and cost of drug development,

with a lower risk of failure.

Drug repurposing has demonstrated success through

serendipitous observations. For example, sildenaﬁl

(Viagra), initially developed for coronary artery dis-

ease, was repurposed for the treatment of erectile dys-

function, reducing development costs and time. Sim-

ilarly, metformin (Glucophage), an oral anti-diabetic

medication, is being developed as a cancer therapeutic

and is currently in phase II/phase III clinical trials.

(Image courtesy:

https://www.researchgate.net/ﬁgure/Methodologies-and-steps-involved-in-drug-

repositioning ﬁg2 342991385)

Figure 2: Traditional drug discovery vs. drug repur-

posing.

1.3 Advantages of Drug Repurposing

Drug repurposing oﬀers several advantages over

traditional drug discovery. It signiﬁcantly reduces the

time spent in research and development (R&D), with an

estimated 3-12 years compared to 10-16 years for tradi-

tional methods. The cost of developing a new drug us-

ing drug repurposing is around $1.6 billion, compared

to $12 billion for conventional methods. Researchers

need only 1-2 years to identify new drug targets and

about eight years to develop a repositioned drug.

Repurposed drugs do not require the initial 6-9

1.4 Methodologies of Drug Repurposing

years typically needed for new drug development but

instead enter directly into preclinical testing and clini-

cal trials, reducing overall risk, time, and cost. Reports

suggest that repurposed drugs require approximately

3-12 years for FDA and European Medicines Agency

(EMA) approval at a 50-60% reduced price. The avail-

ability of pre-clinical and clinical eﬃcacy and safety

information at the start of a repurposing project re-

duces the risks associated with early-stage failures. It

signiﬁcantly lowers costs, increasing clinical safety and

success rates.

(Image courtesy:

https:

//www.google.com/url?sa=i&url=https%3A%2F%2Fsitn.hms.harvard.edu%2Fﬂash%2F2016%

2Fre-engineering-cures-big-data-age-precision-medicine-cocomputational-drug-repositioning%

2F&psig)

Figure 3: Fre-engineering-cures-big-data-age-

Precision-medicine-computational-drug-repositioning

and its illustration.

Cost-Eﬀectiveness: Since repurposed drugs have

already undergone safety testing, the development pro-

cess is often faster and less expensive than new drug

development.

Established Safety Proﬁles: Approved drugs come

with known pharmacokinetics and safety proﬁles, which

can signiﬁcantly reduce the risks associated with new

treatments.

Rapid Application: In urgent situations, such as

the COVID-19 pandemic, repurposing existing drugs

can provide quicker therapeutic options for patients.

Increased Success Rate: Repurposed drugs have

already been proven safe for human use, which reduces

the risk of failure during clinical development. This is

particularly important, as many traditional drug discov-

ery projects fail due to safety issues. The availability

of pre-clinical and clinical data at the start of a re-

purposing project increases the likelihood of success.

This data can help guide the development process and

identify potential issues early on.

Addressing Unmet Medical Needs: Drug repur-

posing can help develop treatments for rare, diﬃcult-

to-treat and neglected diseases. These conditions often

lack eﬀective therapies, and the lower cost and risk as-

sociated with drug repurposing make it an attractive

option for addressing these unmet medical needs.

1.4 Methodologies of Drug

Repurposing

The methodologies adopted in drug repurposing

can be divided into three broad groups based on the

quantity and quality of pharmacological, toxicological,

and biological activity information available:

1. Drug-Oriented Methodology: This approach eval-

uates the structural characteristics, biological ac-

tivities, adverse eﬀects, and toxicities of drug

molecules. It identiﬁes molecules with biolog-

ical eﬀects based on cell/animal assays without

necessarily knowing the biological targets. Suc-

cesses in this methodology include the discovery

of sildenaﬁl through serendipity or clinical ob-

servation.

2. Target-Oriented Methodology: This method in-

volves in silico screening or virtual high-throughput

screening (vHTS) of drugs or compounds from

libraries/databases, such as ligand-based screen-

ing or molecular docking, followed by in vitro

and in vivo high-throughput and high-content

screening (HTS/HCS) against a selective pro-

tein molecule or biomarker. This method has

a higher success rate in drug discovery because

most biological targets directly represent disease

pathways/mechanisms.

3. Disease/Therapy-Oriented Methodology: This

approach is relevant when more information about

the disease model is available. It is guided

by disease and treatment based on proteomics,

genomics, metabolomics, and phenotypic data

concerning the disease process. This method

requires the construction of speciﬁc disease net-

1.4 Methodologies of Drug Repurposing

works, recognising genetic expression, consider-

ing key targets, and identifying disease-causing

protein molecules related to cell and metabolic

pathways of interest in the disease model.

Examples of repurposed drugs

Conclusion

Drug repurposing is an emerging strategy in drug

discovery that oﬀers signiﬁcant advantages over tra-

1.4 Methodologies of Drug Repurposing

ditional methods. By leveraging existing drugs with

known safety proﬁles, drug repurposing reduces the

time, cost, and risk associated with drug development.

Bioinformatics, cheminformatics, and artiﬁcial intelli-

gence advancements have further accelerated the drug

repurposing. This approach is particularly valuable for

developing treatments for rare, diﬃcult-to-treat and ne-

glected diseases. As the ﬁeld continues to evolve, drug

repurposing is poised to play an increasingly important

role in the future of drug discovery. Drug repurposing

(DR), also known as drug repositioning, is a strategic

approach to drug discovery that focuses on identifying

new therapeutic uses for existing drugs. This method

is highly eﬃcient, time-saving, and cost-eﬀective and

minimises the risk of failure, thereby maximising the

therapeutic value of a drug and increasing the success

rate.

In contrast to the traditional drug discovery pro-

cess, which involves de novo identiﬁcation and devel-

opment of new molecular entities (NMEs), drug repur-

posing leverages the combined eﬀorts of activity-based

(experimental) and in silico-based (computational) ap-

proaches to develop new uses for existing drugs. Drug

repurposing represents a promising strategy in drug

discovery, oﬀering a pathway to quickly identify new

therapeutic applications for existing drugs. With ad-

vancements in computational methods and a growing

understanding of disease mechanisms, the potential

for successful drug repositioning continues to expand,

making it a vital area of research in addressing unmet

medical needs.

References

https://www.sciencedirect.com

drug-repurposing-strategies-challenges-and-successes

Illustration-of-the-steps-involved-in-the-drug-repurposing-

process-Step-1-Screening

Fre-engineering-cures-big-data-age-precision-medicine-

computational-drug-repositioning

Illustration-of-the-steps-involved-in-the-drug-repurposing-

process-Step-1-Screening

Kulkarni, V. S., Alagarsamy, V., Solomon, V. R.,

Jose, P. A., & Murugesan, S. (2023). Drug Repur-

posing: An Eﬀective Tool in Modern Drug Discovery.

Russian Journal of Bioorganic Chemistry, 49(2), 157–

166.

https://www.ncbi.nlm.nih.gov/pmc/articles/PMC9945820/

Wang, Y., Aldahdooh, J., Hu, Y. et al. DrugRepo:

a novel approach to repurposing drugs based on chem-

ical and genomic features. Sci Rep 12, 21116 (2022).

About the Author

Geetha Paul is one of the directors of

airis4D. She leads the Biosciences Division. Her

research interests extends from Cell & Molecular Bi-

ology to Environmental Sciences, Odonatology, and

Aquatic Biology.

The Evolution of Artiﬁcial Intelligence in

Neonatology: An India-Focused Perspective

by Kalyani Bagri

airis4D, Vol.2, No.9, 2024

www.airis4d.com

2.1 Introduction

Artiﬁcial Intelligence (AI) is transforming neona-

tology globally, with leading healthcare systems in-

tegrating AI-driven solutions into Neonatal Intensive

Care Units (NICUs). This advancement sets new stan-

dards for monitoring, diagnosing, and treating new-

borns. In contrast, India’s adoption of AI has been

more gradual. However, signiﬁcant progress is being

made with innovative and cost-eﬀective AI solutions

tailored to India’s unique healthcare challenges. In-

creased investment in digital health infrastructure and

AI innovation is expected to accelerate the develop-

ment and adoption of these technologies, narrowing

the gap between India’s advancements and global stan-

dards. This acceleration promises to enhance neonatal

care in India, leading to more widespread use of AI

technologies and improved outcomes for newborns na-

tionwide.

2.2 AI and Neonatology: A Growing

Synergy

AI’s potential in neonatology is profound. AI-

powered tools are increasingly used to predict com-

plications, provide early warnings, and enable timely

interventions. Advanced algorithms analyze medical

images with high precision, aiding in the accurate di-

agnosis of conditions in newborns. These technolo-

gies support clinical decision-making by oﬀering real-

time insights and recommendations crucial for the care

of the most vulnerable infants. By analyzing pat-

terns in neonatal data, AI can forecast complications

such as neonatal sepsis, patent ductus arteriosus, and

other hemodynamic disturbances, as well as short-term

outcomes like mortality, bronchopulmonary dysplasia,

and brain injury. This capability for early interven-

tion is essential for improving neonatal outcomes and

supports the development of personalized treatment

strategies.

Despite these advantages, the adoption of AI in

neonatology in India faces challenges. Fragmented

healthcare infrastructure complicates the integration

of AI solutions across institutions. Data quality is-

sues and resistance from some healthcare providers

also hinder widespread implementation. However, re-

cent advancements in Machine Learning (ML) and

Deep Learning (DL) are driving increased AI adop-

tion. Predictive analytics powered by AI is crucial for

identifying high-risk infants in NICUs, enabling timely

interventions that improve outcomes. AI-driven com-

puter vision enhances the accuracy of diagnosing con-

ditions like respiratory distress syndrome by providing

advanced analysis of medical images, supporting clin-

icians in making more informed decisions.

Machine Learning (ML) algorithms are vital for

predicting neonatal outcomes by analyzing data from

electronic health records, sensors, and monitors. They

play a key role in the early detection of critical con-

2.3 AI in Electronic Neonatal Intensive Care Units (e-NICU)

ditions such as neonatal sepsis, respiratory distress

syndrome, and bronchopulmonary dysplasia, which is

essential for optimizing treatment and improving out-

comes.

Natural Language Processing (NLP) extracts valu-

able insights from clinical notes and reports, improv-

ing care quality and facilitating research. AI-powered

telemedicine platforms expand access to neonatal ex-

pertise, particularly in rural areas, by enabling re-

mote consultations and continuous monitoring. Clini-

cal Decision Support Systems (CDSSs) powered by AI

provide healthcare professionals with real-time, data-

driven recommendations, enhancing decision-making

in neonatal care with precise and timely insights.

AI also fosters collaboration among researchers,

clinicians, and institutions in India. Prominent orga-

nizations and initiatives, including the Indian Academy

of Pediatrics (IAP), National Neonatology Forum (NNF),

All India Institute of Medical Sciences (AIIMS), and

Indian Institutes of Technology (IITs), are spearhead-

ing eﬀorts to implement and integrate AI in neonatol-

ogy. They actively support and advance the use of AI

tools to improve neonatal health outcomes and prac-

tices.

AI solutions in neonatology are designed to com-

plement, rather than replace, the expertise of healthcare

professionals. They assist clinicians by providing data-

driven insights and automating routine tasks, allowing

healthcare professionals to focus on complex decision-

making and personalized care. For example, AI-driven

systems can manage routine monitoring tasks, enabling

professionals to focus on interpreting complex data and

making informed decisions to improve neonatal out-

comes.

2.3 AI in Electronic Neonatal

Intensive Care Units (e-NICU)

AI is revolutionizing Neonatal Intensive Care Units

(e-NICUs) in India by signiﬁcantly improving patient

care and operational eﬃciency. As the incidence of

preterm births and neonatal complications rises, AI

technologies are increasingly integrated to enhance

clinical outcomes and streamline e-NICU operations.

These systems facilitate continuous monitoring of vital

signs, early detection of complications, and personal-

ized treatment plans through the analysis of data from

electronic health records and real-time devices. AI-

driven decision support systems assist clinicians with

evidence-based recommendations, improving clinical

judgment and ensuring timely, accurate care.

The integration of AI into e-NICU in India is

driven by dynamic collaboration among government

institutions, private healthcare providers, non-proﬁts,

and technology companies. Notable initiatives such as

the National Health Mission’s e-NICU program, AI-

IMS’ TeleNICU platform, Apollo Hospitals’ e-NICU

initiative, and the Medtronic-India e-NICU program

highlight a uniﬁed commitment to revolutionizing neona-

tal care. These eﬀorts reﬂect a concerted push to lever-

age AI for enhancing patient outcomes and streamlin-

ing operations. They also extend specialized care to

underserved regions, demonstrating a robust and mul-

tifaceted approach to advancing neonatal healthcare in

India. Support from organizations such as the Min-

istry of Health and Family Welfare, Save the Children,

and UNICEF signiﬁcantly ampliﬁes these initiatives.

Their backing enhances eﬀorts to improve monitoring,

predict adverse events, and personalize care, ensur-

ing equitable access to high-quality neonatal services

throughout India, regardless of geographic location.

References

1. Keles, E., Bagci, U. The past, current, and fu-

ture of neonatal intensive care units with artiﬁ-

cial intelligence: a systematic review. npj Digit.

2.3 AI in Electronic Neonatal Intensive Care Units (e-NICU)

Med. 6, 220 (2023). https://doi.org/10.1038/

s41746-023-00941-5

2. Beam, K., Sharma, P., Levy, P. et al. Artiﬁcial

intelligence in the neonatal intensive care unit:

the time is now. J Perinatol 44, 131–135 (2024).

https://doi.org/10.1038/s41372-023-01719-z

3. Chioma, R.; Sbordone, A.; Patti, M.L.; Perri, A.;

Vento, G.; Nobile, S. Applications of Artiﬁcial

Intelligence in Neonatology. Appl. Sci. 2023,

13, 3211. https://doi.org/10.3390/app13053211

4. Patel, S.Y., Palma, J.P., Hoﬀman, J.M. et al.

Neonatal informatics: past, present and future.

J Perinatol 44, 773–776 (2024). https://doi.org/

10.1038/s41372-024-01924-4

5. Rallis, D.; Baltogianni, M.; Kapetaniou, K.; Gi-

apros, V. Current Applications of Artiﬁcial In-

telligence in the Neonatal Intensive Care Unit.

BioMedInformatics 2024, 4, 1225-1248. https:

//doi.org/10.3390/biomedinformatics402006

6. Sullivan, B.A., Beam, K., Vesoulis, Z.A., Aziz,

K.B., Husain, A.N., Knake, L.A., Moreira, A.G.,

Hooven, T.A., Weiss, E.M., Carr, N.R., El-Ferzli,

G.T., Patel, R.M., Simek, K.A., Hernandez, A.J.,

Barry, J.S., McAdams, R.M. Transforming neona-

tal care with artiﬁcial intelligence: challenges,

ethical consideration, and opportunities. J Peri-

natol. 2024 Jan; 44(1):1-11. https://doi.org/10.

1038/s41372-023-01848-5

7. Yang, X., Zhang, J., Cao, M., Pan, Y., Zhang,

Y. Application of e-health on neonatal inten-

sive care unit discharged preterm infants and

their parents: Protocol for systematic review

and meta-analysis. Digit Health. 2023 Oct

9; 9:20552076231205271. https://doi.org/10.

1177/20552076231205271

About the Author

Dr. Kalyani Bagri is a Senior Research

Associate at Fernandez Foundation in Hyderabad, In-

dia, where she plays a pivotal role as the lead data sci-

entist in the Neonatology Department. She earned her

Ph.D. in Astrophysics from Pt. Ravishankar Shukla

University, in collaboration with IUCAA and TIFR

Mumbai. In her current role, Dr. Bagri independently

integrates Artiﬁcial Intelligence (AI) into neonatal

care. She is leading critical projects, including the

development of a Bronchopulmonary Dysplasia esti-

mator to guide strategic decisions and optimize steroid

usage in neonates, and a sepsis calculator designed to

detect sepsis in neonates prior to clinical recognition.

Additionally, she is involved in a range of initiatives

aimed at advancing neonatal health outcomes through

her innovative and data-driven approaches.

Biomarkers: Powerful Indicators in

Healthcare

by Jinsu Ann Mathew

airis4D, Vol.2, No.9, 2024

www.airis4d.com

Biomarkers are essential tools in modern medicine

that help doctors and scientists understand what’s hap-

pening inside our bodies. A biomarker is any measur-

able indicator of a biological condition or process, such

as a molecule in our blood, a gene, or even a physical

measurement like blood pressure. These indicators are

used to diagnose diseases, predict health risks, moni-

tor the eﬀects of treatments, and guide decisions on the

best therapies for individual patients.

For example, cholesterol levels are biomarkers

that help assess the risk of heart disease. In cancer

treatment, speciﬁc genetic biomarkers can determine

whether a patient will respond well to a particular ther-

apy. By understanding these indicators, doctors can

tailor treatments to ﬁt each patient’s needs, leading to

better health outcomes.

With advances in science and technology, diﬀer-

ent types of biomarkers are being identiﬁed and cat-

egorized based on their speciﬁc roles in healthcare.

Some biomarkers are used to diagnose diseases, while

others predict a person’s risk of developing a condi-

tion, monitor the progress of a disease, or assess how

well a treatment is working. This article will explore

the various types of biomarkers, including diagnos-

tic, prognostic, predictive, risk, safety, and monitor-

ing biomarkers, and explain how each of them plays

a unique role in improving patient care and guiding

medical decisions.

3.1 Diagnostic Biomarkers

Diagnostic biomarkers are critical tools in modern

medicine, used to detect and conﬁrm the presence of a

speciﬁc disease or medical condition. These biomark-

ers are measurable indicators—often proteins, genes,

or other molecules—that can be detected in bodily ﬂu-

ids, tissues, or other samples. Their primary function is

to aid in the accurate and timely diagnosis of diseases,

which is essential for initiating appropriate treatment

and improving patient outcomes. For example, the

detection of human chorionic gonadotropin (hCG) in

urine is a well-known diagnostic biomarker used in

pregnancy tests, oﬀering a quick and accurate method

to conﬁrm pregnancy.

The application of diagnostic biomarkers spans a

wide array of medical ﬁelds, from oncology and car-

diology to infectious diseases and neurological disor-

ders. In oncology, for instance, the presence of cancer-

speciﬁc biomarkers like CA-125 for ovarian cancer or

carcinoembryonic antigen (CEA) for colorectal can-

cer helps clinicians detect malignancies at an earlier

stage, often before symptoms arise. Early detection

through these biomarkers can lead to timely interven-

tion, improving the chances of successful treatment and

survival. Similarly, in infectious diseases, diagnostic

biomarkers such as viral RNA or speciﬁc antibodies

allow for rapid identiﬁcation of pathogens, enabling

targeted treatment and reducing the spread of infec-

tion.

3.2 Prognostic Biomarkers

Unlike diagnostic biomarkers, which are used to

detect the presence of a disease, prognostic biomarkers

help predict the future progression of the disease, oﬀer-

ing insights into the expected clinical trajectory. These

biomarkers can indicate whether a disease is likely to

progress rapidly or remain stable, whether a patient

is at higher risk of complications, or what the overall

chances of survival might be. For example, in breast

cancer, the overexpression of the HER2 gene is a prog-

nostic biomarker that indicates a more aggressive form

of the disease, often associated with a poorer outcome.

Prognostic biomarkers are particularly useful in

stratifying patients based on their risk levels, allowing

healthcare providers to tailor treatment plans accord-

ing to the severity and expected progression of the

disease. In oncology, for instance, the presence of

speciﬁc genetic mutations, such as the BRAF muta-

tion in melanoma, can signal a more aggressive dis-

ease course, guiding decisions about the intensity and

type of treatment needed. Similarly, in cardiovascu-

lar disease, elevated levels of certain biomarkers, like

NT-proBNP, can predict a higher risk of heart fail-

ure, prompting closer monitoring and more aggressive

management to prevent adverse outcomes.

The ability of prognostic biomarkers to predict

disease outcomes makes them invaluable in both clin-

ical practice and research. They help clinicians decide

on the best course of action, whether it involves more

aggressive treatment for high-risk patients or a more

conservative approach for those with a favorable prog-

nosis. Additionally, prognostic biomarkers play a key

role in clinical trials, where they are used to identify

patient subgroups that might beneﬁt most from a new

therapy, thereby enhancing the precision and eﬃciency

of clinical research.

3.3 Predictive Biomarkers

Predictive biomarkers are integral to the advance-

ment of personalized medicine, oﬀering a way to an-

ticipate how an individual patient is likely to respond

to a speciﬁc treatment. These biomarkers are mea-

surable indicators, such as genetic variations, protein

levels, or other molecular features, that can forecast the

eﬀectiveness or potential side eﬀects of a therapeutic

intervention before it is administered. By identifying

patients who are more likely to beneﬁt from a particu-

lar treatment or who may experience adverse reactions,

predictive biomarkers allow healthcare providers to tai-

lor therapies to the unique needs of each patient. This

customization not only enhances treatment eﬃcacy but

also minimizes the risk of unnecessary side eﬀects.

The application of predictive biomarkers is broad,

inﬂuencing treatment decisions across various medical

ﬁelds. They help clinicians choose the most appropri-

ate therapy for each patient, avoiding a one-size-ﬁts-all

approach. For instance, certain biomarkers might indi-

cate that a patient is likely to respond well to a speciﬁc

drug, guiding the selection of that treatment. Con-

versely, the presence or absence of a biomarker might

suggest that a diﬀerent therapy would be more eﬀec-

tive, helping to avoid treatments that are unlikely to

work. This approach leads to more targeted and eﬃ-

cient care, reducing the trial-and-error often associated

with treatment decisions.

Moreover, predictive biomarkers are not just about

selecting the right medication. They also provide in-

sights into the appropriate dosage, frequency, and du-

ration of treatment, further reﬁning patient care. By

understanding a patient’s unique biological character-

istics, healthcare providers can better predict how a

treatment will interact with their body, leading to more

precise and individualized care. In summary, predic-

tive biomarkers are essential tools in modern medicine,

enabling healthcare providers to personalize treatment

strategies, improve patient outcomes, and ensure that

each patient receives the most eﬀective and safest care

possible.

3.4 Risk Biomarkers

Risk biomarkers are biological indicators that help

predict an individual’s likelihood of developing a spe-

ciﬁc disease or medical condition in the future. Unlike

diagnostic biomarkers, which detect the presence of

an existing disease, risk biomarkers are used in pre-

3.5 Safety Biomarkers

ventive medicine to identify individuals who are at an

increased risk of developing a condition before any

symptoms or clinical signs appear. These biomarkers

provide valuable insights into a person’s health, en-

abling early interventions that can potentially delay or

prevent the onset of disease.

Risk biomarkers can be genetic, biochemical, or

physiological in nature. For instance, certain genetic

mutations, like those found in the BRCA1 and BRCA2

genes, are well-known risk biomarkers for breast and

ovarian cancer. Individuals who carry these muta-

tions have a signiﬁcantly higher lifetime risk of devel-

oping these cancers compared to the general popula-

tion. Knowing this information allows for personalized

preventive strategies, such as increased surveillance,

lifestyle modiﬁcations, or even prophylactic surgeries,

to reduce the risk of cancer development.

The utility of risk biomarkers lies in their ability

to stratify populations based on their susceptibility to

disease, allowing healthcare providers to focus preven-

tive eﬀorts on those who are most at risk. This targeted

approach not only improves health outcomes by pre-

venting or delaying disease onset but also contributes

to more eﬃcient use of healthcare resources. In sum-

mary, risk biomarkers play a critical role in predictive

and preventive medicine, oﬀering a proactive means of

managing health and reducing the burden of chronic

diseases.

3.5 Safety Biomarkers

Safety biomarkers are specialized indicators used

in medicine and clinical research to detect, predict, or

monitor the potential adverse eﬀects of a drug, treat-

ment, or environmental exposure. These biomarkers

are crucial in ensuring that therapies are safe for pa-

tients by identifying early signs of toxicity or other

harmful eﬀects before they become clinically signif-

icant. Their primary role is to safeguard patients

during treatment by providing real-time information

about how a body responds to a particular interven-

tion. In clinical practice and drug development, safety

biomarkers play a pivotal role in identifying early signs

of harm. They help detect physiological changes that

might indicate stress or damage to speciﬁc organs or

systems in the body. For example, if a biomarker shows

an abnormal increase in a speciﬁc measurement, it

could signal that a particular organ is under strain due

to the treatment. This allows healthcare providers to

intervene promptly, potentially preventing more severe

side eﬀects or long-term damage.

During the development of new drugs or treat-

ments, safety biomarkers are used extensively to eval-

uate the tolerability of a therapy. They provide critical

data on how the body reacts at diﬀerent dosage levels,

helping to identify the safest and most eﬀective dose.

This information is vital in clinical trials, where ensur-

ing participant safety is paramount. If safety biomark-

ers indicate potential risks, researchers can adjust the

treatment protocol or halt the trial to prevent harm to

participants.

In ongoing patient care, safety biomarkers are

used to monitor the eﬀects of long-term treatments.

They provide continuous feedback on whether a treat-

ment remains safe for the patient over time. This on-

going monitoring helps manage the delicate balance

between treatment eﬃcacy and safety, ensuring that

the beneﬁts of a therapy outweigh any potential risks.

By detecting adverse eﬀects early, safety biomarkers

enable timely interventions, which can include alter-

ing the treatment regimen or implementing additional

protective measures.

3.6 Monitoring Biomarkers

Monitoring biomarkers are used to track the pro-

gression of a disease or to assess the eﬀectiveness of

a treatment over time. They provide ongoing informa-

tion about the status of a condition or the impact of a

therapeutic intervention. These biomarkers are often

measured repeatedly during the course of treatment to

guide clinical decisions, adjust therapies, and predict

outcomes.

One of the key applications of monitoring biomark-

ers is in chronic diseases, where regular assessment is

crucial for managing long-term health. For example,

in diabetes management, Hemoglobin A1c (HbA1c) is

a widely used monitoring biomarker that reﬂects av-

3.7 Summary of Biomarker Types

erage blood glucose levels over the past two to three

months. By regularly measuring HbA1c, healthcare

providers can determine how well a patient’s diabetes

is being controlled and whether adjustments in medi-

cation, diet, or lifestyle are needed. Similarly, in cancer

treatment, the levels of tumor markers such as prostate-

speciﬁc antigen (PSA) in prostate cancer or CA-125 in

ovarian cancer are monitored to evaluate the eﬀective-

ness of therapy and to detect potential recurrences at

an early stage.

Monitoring biomarkers are also essential in man-

aging infectious diseases. In HIV treatment, for in-

stance, the viral load—the amount of HIV RNA in the

blood—is a critical monitoring biomarker used to as-

sess how eﬀectively antiretroviral therapy is suppress-

ing the virus. A declining viral load indicates that the

treatment is working, while an increase may suggest

drug resistance or poor adherence to therapy, prompt-

ing a change in treatment strategy.

The utility of monitoring biomarkers lies in their

ability to provide dynamic, real-time information about

a patient’s health status. This continuous feedback al-

lows for more personalized and adaptive healthcare,

where treatments can be ﬁne-tuned based on the pa-

tient’s response, thereby optimizing outcomes. Fur-

thermore, monitoring biomarkers can help in predict-

ing relapses or complications, enabling preemptive in-

terventions that can improve prognosis and quality of

life. In summary, monitoring biomarkers are essen-

tial in the ongoing management of diseases, oﬀering

critical insights into treatment eﬃcacy and disease pro-

gression, and ultimately guiding more precise and re-

sponsive patient care.

3.7 Summary of Biomarker Types

In this article, we have explored the diﬀerent types

of biomarkers and their critical roles in medical diag-

nosis, treatment, and monitoring. To provide a clear

and concise overview, Figure ?? summarizes the main

functions of each biomarker type. This ﬁgure is de-

signed to encapsulate the essence of each biomarker’s

purpose, aiding in the quick understanding and diﬀer-

entiation of their roles.

Figure 1: Diﬀerent types of biomarkers

3.8 Conclusion

In conclusion, biomarkers play an essential role

in modern medicine by providing valuable informa-

tion across various stages of disease management. Di-

agnostic biomarkers enable precise identiﬁcation of

diseases, while prognostic biomarkers oﬀer insights

into disease severity and potential outcomes. Predic-

tive biomarkers help tailor treatment strategies to in-

dividual patients, ensuring that therapies are eﬀective.

Risk biomarkers assess the likelihood of disease de-

velopment, allowing for preventive measures. Safety

biomarkers monitor potential adverse eﬀects of treat-

ments, safeguarding patient health. Finally, monitoring

biomarkers track the progression or response to treat-

ment, guiding ongoing therapeutic decisions.

Understanding the distinct functions of these biomark-

ers enhances their application in clinical practice, lead-

ing to more eﬀective and personalized healthcare. By

integrating these biomarkers into routine practice, clin-

icians can improve diagnostic accuracy, optimize treat-

ment regimens, and ultimately enhance patient out-

comes.

References

An Introduction to Biomarker Types & Applica-

tions

Biomarkers and Their Role in Detection of Biomolecules

Clinical trials, how biomarkers help research

What are Biomarkers? What are the types of

Biomarkers?

Cancer biomarker

Biomarker discovery using scRNA-seq: what’s

the big deal?

3.8 Conclusion

About the Author

Jinsu Ann Mathew is a research scholar

in Natural Language Processing and Chemical Infor-

matics. Her interests include applying basic scientiﬁc

research on computational linguistics, practical appli-

cations of human language technology, and interdis-

ciplinary work in computational physics.

Part IV

General

The Swachha Shala Initiative

by Chinar Deshpande

airis4D, Vol.2, No.9, 2024

www.airis4d.com

The Swachha Shala Initiative is a project aimed

towards helping students from all corners of India get

access to clean and hygienic bathrooms, especially in

schools. According to UNICEF (1), if bathrooms are

not clean, students (mainly girls) are much less likely to

use them, leading to repeated absences from schools,

and in some cases, even dropouts. Swachha means

clean, and Shala means school in Marathi. Putting

them together gives us Swachha Shala, but the project

is so much bigger than making schools clean. Swachha

also refers to purity. Through this initiative, I aim to

transform schools into a safe haven for all students to

learn and grow as individuals. At a high level, the goal

for this initiative is to make sure that all the young boys

and girls of the current generation, and future ones, are

able to learn unhindered of all obstacles.

1.1 The Problem

The presence of clean bathrooms in schools is

directly related to the attendance of girls in the school.

This is due to multiple reasons, but the primary one

relates to menstruation.

If the bathrooms are not clean, girls will not be

able to manage menstruation well in schools. This

leads to reluctance in girls to use these facilities, and

this leads to their absence during menstrual cycles.

Since there is also a stigma surrounding menstruation

in most cultures, girls are forced to face insults and

bullying if they are not able to manage menstruation in

the school. Overall, this leads to a negative connotation

being developed in the minds of girls, leading to lower

academic performance, non-attendance, and dropouts.

Figure 1: The workings of a CNN

1.2 My Approach

In the research space, there are multiple articles

and research papers published about Dirt Detection

Systems and Street Litter Detection. These papers were

invaluable towards ﬁguring out a suitable approach.

Since these research papers are similar to the area of

Computer Vision explored by this project, they were

a part of the literature review conducted. There were

some interesting ﬁndings collected from these papers.

Canedo et al (2) proposed a YOLOv5 driven ap-

proach, utilizing Transfer Learning to train an ML

model on a 20,000 image dataset. Although achieving

only a 73.3% accuracy, Canedo et al put forth a compre-

hensive algorithm for such types of applications which

is planned to be incorporated in the project.

The main solution ideated was to use a Convo-

lutional Neural Network (illustrated in Figure 1) to

classify images as clean or dirty, then to contact the

schools where immediate attention was required. One

interesting optimizer that can be helpful to the CNN

in this case is the Stochastic Gradient Descent (SGD)

to ensure a smooth learning curve. However, the main

obstacle was collecting these images.

1.3 Dataset

Figure 2: Ping et al’s Object Detection Algorithm

To solve this issue, Ping et al (3) were invalu-

able. In their paper about Road Litter Detection, they

collected images of the streets of San Jose using their

students. Also, since they were using IOU and Ob-

ject Detection (see Figure 2) to identify various types

of objects in the road, this project adopted a speciﬁc

strategy: the students could upload images of the bath-

rooms themselves, without having to rely on any third

parties. These images could be collected and the algo-

rithm could be run on them, predicting whether each

image was of a clean or dirty bathroom. Here, al-

gorithms like Object Detection and Detection Boxes

could be used to identify common indicators of a clean/

dirty bathroom. If there were multiple occurrences of

the same bathroom being ﬂagged by diﬀerent students

over a consistent period of time, the school would be

pinpointed and reached out to.

1.3 Dataset

For any Neural Network, a suitable dataset is re-

quired to train the model. In this case, since the project

was a very niche application of ML, there were no

datasets which were varied and detailed enough for the

model to be trained. To solve this issue, an in-depth

literature review was conducted to try to discover any

images/ datasets that could be used for the model train-

ing.

Although a comprehensive bathroom hygiene de-

tection dataset was not found during the literature re-

view, the impurities in a conventional bathroom could

be classiﬁed under categories such as dirt, trash, litter,

etc. All these are sub-categories in datasets such as

ImageNet, which can easily be accessed. Additionally,

there are a few datasets such as ACIN speciﬁcally for

dirt, that can be used for a more in-depth training pro-

cess. Lastly, there is also the option of web scraping

for images online (with the necessary permissions), to

generate a sizable dataset for some of the more uncom-

mon categories.

1.4 Current App

The current app is a React Native App called

Swachha Shala. At the moment, it is in the open testing

stage on the Play Store. There are plans to upload it to

the App Store as well in the future.

The app is a simple one, as mentioned above. It

simply collects images from the users, along with the

classiﬁcation of each image into the categories of clean

and dirty. Additionally, the app also pinpoints exactly

where in India the image is being uploaded from. It

also has a text input for the user to enter the school

name, which is all important data used to store each

recording in an automatic Google Drive directory.

This app stores the coordinates of the upload, the

image itself, and the classiﬁcation data in Google Fire-

base. Once the app is out on the Play Store, the data

collection process will begin. A CNN will be built to

make sure that the prediction capabilities of the overall

algorithm are as accurate as possible, to allow for the

performance of the app to be as smooth as possible.

There are also plans to incorporate a more detailed

veriﬁcation system within the app. Template Match-

ing and Non-Max Suppression will be used to conﬁrm

the identity of each student, and check that the data

provided is accurate.

1.5 Epilogue

Currently, the process of model training is to start.

Once the model is ready, the true capabilities of the Ini-

tiative will come out. The model and algorithm will

need a few tweaks and changes, but the goal is for stu-

dents to be involved in the hygiene and cleanliness of

their school, helping it with their contributions. Hope-

fully, The Swachha Shala Initiative will be a beacon of

1.5 Epilogue

Figure 3: Complete Process of uploading an Image

light for all the young students of India, illuminating

the path towards a better education, and a prosperous

future.

References

1. https://towardsdatascience.com/a-comprehensive-

guide-to-convolutional-neural-networks-the-eli5-

way-3bd2b1164a53

2. https://www.mdpi.com/2227-7080/9/4/94

3. https://www.researchgate.net/publication/341182644

4. https://github.com/nuwandda/clean-littered-road-

classiﬁcation

5. https://www.thehindu.com/opinion/op-ed/the-link-

between-sanitation-and-schooling/article6423571.ece

About the Author

Chinar Deshpande is a 12th Grader study-

ing in Aditya English Medium School. He is pas-

sionate about research, having worked on projects in

Computer Vision, Machine Learning, and the AI x Ed-

ucation Space. Additionally, he has also completed in-

ternships at companies such as EagleRobotLab, Ban-

galore, which aims to build humanoid robots capable

of teaching and assisting instructors in the classroom.

Lastly, he is also collaborating with Pune Knowl-

edge Cluster, having demonstrated multiple POCs and

projects to the team.

Part V

Computer Programming

The Evolution of GPUs: Powering the Next

Generation of High-End Computing and

Artiﬁcial Intelligence

by Atharva Pathak

airis4D, Vol.2, No.9, 2024

www.airis4d.com

1.1 Introduction

Graphics processing units, or GPUs, have come a

long way from their early days as specialized hardware

for rendering video game graphics. Over the last two

decades, these powerful processors have evolved into

the backbone of high-end computing, driving advances

in ﬁelds ranging from scientiﬁc research to artiﬁcial in-

telligence (AI) and machine learning (ML). This arti-

cle explores the fascinating journey of GPU evolution,

their architectural innovations, and how they have be-

come indispensable tools in modern computing.

1.2 The Birth of the GPU: From

Graphics to General-Purpose

Processing

The GPU was born out of necessity. In the early

days of personal computing, CPUs were the workhorses

of the system, handling everything from calculations

to graphics rendering. However, as video games and

3D graphics became more sophisticated in the 1990s,

CPUs struggled to keep up with the demand for real-

time processing of complex visual data. This led to the

development of the ﬁrst GPUs, which were designed

to oﬄoad these tasks from the CPU.

NVIDIA’s introduction of the GeForce 256 in

1999 marked a signiﬁcant milestone. This was the ﬁrst

chip to be marketed as a graphics processing unit, capa-

ble of handling millions of polygons per second. Its ar-

chitecture was designed to perform many calculations

simultaneously, or in parallel, which was perfect for

rendering complex 3D scenes quickly and eﬃciently.

Unlike CPUs, which are optimized for sequential tasks,

GPUs excel at parallel processing, making them ideal

for graphics-intensive applications.

1.3 GPU Architecture: A Deep Dive

At the core of a GPU’s power is its architecture,

which has evolved dramatically since the early days.

Early GPUs like the GeForce 256 were focused solely

on graphics, with ﬁxed-function pipelines that limited

their ﬂexibility. These pipelines consisted of several

stages (vertex processing, rasterization, pixel process-

ing) that were hardwired to perform speciﬁc tasks.

However, as the demand for more ﬂexible and

powerful GPUs grew, manufacturers began to develop

programmable pipelines. This shift was evident in

NVIDIA’s G80 architecture, introduced in 2006, which

featured uniﬁed shader cores. These cores could per-

form both vertex and pixel processing tasks, making

the GPU much more versatile. This architectural in-

novation laid the groundwork for the GPU’s expansion

into general-purpose computing.

Modern GPUs, such as those in NVIDIA’s Tesla

series, feature thousands of CUDA cores, each capable

1.4 From Graphics to General-Purpose Computing

of executing threads independently. These cores are

grouped into streaming multiprocessors (SMs), which

can execute multiple warps (groups of threads) simulta-

neously. This architecture allows GPUs to handle mas-

sive amounts of data in parallel, making them ideal for

high-performance computing (HPC) tasks such as sci-

entiﬁc simulations, big data analytics, and, of course,

AI and ML.

1.4 From Graphics to

General-Purpose Computing

The shift from ﬁxed-function pipelines to pro-

grammable shaders in GPUs was a game-changer. It

opened the door to using GPUs for tasks beyond graph-

ics, leading to the concept of general-purpose comput-

ing on graphics processing units (GPGPU). This ap-

proach leverages the massive parallel processing power

of GPUs for a wide range of applications.

A pivotal moment in the evolution of GPGPU was

the introduction of CUDA (Compute Uniﬁed Device

Architecture) by NVIDIA in 2007. CUDA allowed de-

velopers to write programs that could run on GPUs,

taking advantage of their parallel processing capabil-

ities. This led to the widespread adoption of GPUs

in ﬁelds like scientiﬁc computing, where tasks such as

simulating physical systems, processing large datasets,

and performing complex calculations could be done

much faster than with traditional CPUs.

For example, in the ﬁeld of genomics, researchers

use GPUs to accelerate the analysis of DNA sequences.

This involves comparing millions of genetic sequences

to identify variations and mutations, a task that would

take weeks or even months on CPUs but can be com-

pleted in a fraction of the time with GPUs.

1.5 GPUs in AI and Machine

Learning: A Match Made in

Silicon

The rise of AI and ML has been one of the most

signiﬁcant technological trends of the past decade, and

GPUs have played a crucial role in this revolution.

Training AI models, particularly deep learning models,

involves adjusting millions or even billions of param-

eters. This requires an enormous amount of compu-

tational power, which GPUs are uniquely equipped to

provide.

The architecture of deep neural networks (DNNs),

which are the foundation of many AI models, involves

multiple layers of interconnected nodes (neurons). Train-

ing these networks requires performing vast amounts of

matrix multiplications, a task that is inherently parallel

and perfectly suited to GPUs. As a result, GPUs have

become the standard for training AI models, reducing

training times from weeks to days or even hours.

One of the most signiﬁcant examples of GPU-

powered AI is NVIDIA’s Tesla V100, built on the Volta

architecture. This GPU features Tensor Cores, special-

ized units designed to accelerate the mixed-precision

matrix multiply-accumulate operations that are at the

heart of deep learning. With Tensor Cores, the Tesla

V100 can deliver up to 125 teraﬂops of performance,

making it one of the most powerful GPUs for AI train-

ing.

This computational power has enabled the devel-

opment of state-of-the-art AI models like OpenAI’s

GPT-4 and Google’s BERT. GPT-4, for example, is a

large language model with billions of parameters, ca-

pable of generating human-like text based on a given

prompt. Training such a model requires massive amounts

of data and computational resources, typically involv-

ing hundreds of GPUs working in parallel for several

days or weeks.

1.6 Challenges and Limitations of

GPUs in High-End Computing

While GPUs have revolutionized high-end com-

puting, they are not without challenges. One of the pri-

mary limitations is power consumption. High-performance

GPUs, such as those used in data centers, can consume

hundreds of watts of power, leading to signiﬁcant en-

ergy costs and heat dissipation challenges. This has

spurred research into more energy-eﬃcient GPU de-

signs and cooling technologies.

1.7 The Future of GPUs: What’s Next?

Another challenge is the complexity of program-

ming for GPUs. While frameworks like CUDA have

made it easier to develop GPU-accelerated applica-

tions, writing eﬃcient GPU code still requires a deep

understanding of parallel programming concepts. This

learning curve can be a barrier for developers who are

accustomed to CPU programming.

Additionally, while GPUs excel at parallel pro-

cessing, they are not as well-suited for tasks that re-

quire high single-thread performance or involve a lot

of branching and decision-making. For these types of

tasks, CPUs still have the edge. As a result, many high-

performance computing systems use a combination of

CPUs and GPUs, each handling the tasks they are best

suited for.

1.7 The Future of GPUs: What’s

Next?

The future of GPUs is closely tied to the contin-

ued evolution of AI and high-performance computing.

As AI models become more complex, the demand for

more powerful and eﬃcient GPUs will only increase.

GPU manufacturers like NVIDIA, AMD, and Intel are

already working on next-generation architectures that

promise even greater performance and eﬃciency.

One of the most exciting developments is the inte-

gration of AI-speciﬁc hardware into GPUs. For exam-

ple, NVIDIA’s latest GPUs feature Tensor Cores, which

are speciﬁcally designed for AI workloads. AMD and

Intel are also developing their own AI-focused tech-

nologies, which will likely lead to even more powerful

and specialized GPUs in the coming years.

Moreover, the convergence of AI with other emerg-

ing technologies, such as quantum computing and neu-

romorphic computing, could lead to new types of GPUs

that are optimized for these applications. Quantum

computing, in particular, has the potential to revolu-

tionize the ﬁeld of AI, and GPUs will likely play a key

role in this new era of computing.

(https://tatourian.blog/2013/09/03/nvidia-gpu-architecture-cuda-programming-environment/)

(https:

//exittechnologies.com/blog/data-center/gpu-technology-gpu-history-and- gpu-prices-over-time/)

1.8 Conclusion

The evolution of GPUs from simple graphics ac-

celerators to the powerhouses of modern computing

is a testament to the relentless pace of technological

innovation. Their ability to perform massive amounts

of parallel computations has made them indispensable

tools in ﬁelds ranging from scientiﬁc research to AI

and ML. As we look to the future, GPUs will continue

to drive the next generation of high-performance com-

puting, enabling new breakthroughs and shaping the

world we live in.

1.9 Suggested Further Reading

Deep Learning with GPU Acceleration by Ian

Goodfellow

The Hardware Behind Artiﬁcial Intelligence by

Jeﬀrey Dean

High-Performance Computing with GPUs: A

Complete Guide by Jack Dongarra

Image Suggestions

1.10 Timeline of GPU Evolution

A detailed timeline showing the key milestones

in the development of GPUs, from early ﬁxed-function

pipelines to modern programmable architectures.

1.11 GPU Architecture Overview

(https://developer.nvidia.com/blog/nvidia-turing-architecture-in- depth/)

(https://www.nature.com/articles/s41598-021-91878-w)

1.11 GPU Architecture Overview

A diagram illustrating the architecture of a mod-

ern GPU, highlighting components like CUDA cores,

Tensor Cores, and streaming multiprocessors.

1.12 GPUs in Action

Visualisations of real-world applications of GPUs

in AI, such as the training of neural networks and the

simulation of scientiﬁc phenomena.

1.13 Future of GPUs

Conceptual images depicting next-generation GPU

architectures, potentially showing the integration of

AI-speciﬁc hardware and quantum computing elements.

This article aims to provide a comprehensive yet

accessible overview of GPUs, highlighting their techni-

cal evolution, current applications, and future potential

in driving high-end computing and AI.

(https:

//www.nextplatform.com/2024/06/02/nvidia-unfolds-gpu-interconnect-roadmaps-out-to-2027/)

(https://towardsdatascience.com/ten-years-of-ai-in- review-85decdb2a540)

About the Author

Atharva Pathak currently work as a Soft-

ware Engineer & Data Manager for the Pune Knowl-

edge Cluster, A project under the Oﬃce of Principal

Scientiﬁc Advisor, Govt. of India & Supported by

IUCAA, Pune, IN. Before this, I was an Astronomer

at the Inter-University Centre for Astronomy & Astro-

physics, IUCAA. I have also worked on various free-

lance projects, development required for websites and

applications, And localization of diﬀerent software.

I am also a life member of Jyotirvidya Parisanstha,

India’s Oldest association of Amateur Astronomers,

and I look after the IOTA-India Occultation section

as a webmaster and data curator.

References

About airis4D

Artiﬁcial Intelligence Research and Intelligent Systems (airis4D) is an AI and Bio-sciences Research Centre.

The Centre aims to create new knowledge in the ﬁeld of Space Science, Astronomy, Robotics, Agri Science,

Industry, and Biodiversity to bring Progress and Plenitude to the People and the Planet.

Vision

Humanity is in the 4th Industrial Revolution era, which operates on a cyber-physical production system. Cutting-

edge research and development in science and technology to create new knowledge and skills become the key to

the new world economy. Most of the resources for this goal can be harnessed by integrating biological systems

with intelligent computing systems oﬀered by AI. The future survival of humans, animals, and the ecosystem

depends on how eﬃciently the realities and resources are responsibly used for abundance and wellness. Artiﬁcial

intelligence Research and Intelligent Systems pursue this vision and look for the best actions that ensure an

abundant environment and ecosystem for the planet and the people.

Mission Statement

The 4D in airis4D represents the mission to Dream, Design, Develop, and Deploy Knowledge with the ﬁre of

commitment and dedication towards humanity and the ecosystem.

Dream

To promote the unlimited human potential to dream the impossible.

Design

To nurture the human capacity to articulate a dream and logically realise it.

Develop

To assist the talents to materialise a design into a product, a service, a knowledge that beneﬁts the community

and the planet.

Deploy

To realise and educate humanity that a knowledge that is not deployed makes no diﬀerence by its absence.

Campus

Situated in a lush green village campus in Thelliyoor, Kerala, India, airis4D was established under the auspicious

of SEED Foundation (Susthiratha, Environment, Education Development Foundation) a not-for-proﬁt company

for promoting Education, Research. Engineering, Biology, Development, etc.

The whole campus is powered by Solar power and has a rain harvesting facility to provide suﬃcient water supply

for up to three months of drought. The computing facility in the campus is accessible from anywhere through a

dedicated optical ﬁbre internet connectivity 24×7.

There is a freshwater stream that originates from the nearby hills and ﬂows through the middle of the campus.

The campus is a noted habitat for the biodiversity of tropical Fauna and Flora. airis4D carry out periodic and

systematic water quality and species diversity surveys in the region to ensure its richness. It is our pride that

the site has consistently been environment-friendly and rich in biodiversity. airis4D is also growing fruit plants

that can feed birds and provide water bodies to survive the drought.