Shasanka Shekhar Padhi

About Me

A Bioinformatics professional with 3+ years of experience in AI/ML model development, biomarker identification, and knowledge graph applications. Skilled in GenAI, LLM, knowledge graph analytics and core bioinformatics solutions to drive data driven insights, enhance decision making, and deliver scalable solutions through cross functional collaboration across diverse domains. Focused on advancing computational research that integrates AI and data-driven science to solve complex problems across biology, healthcare, and beyond.

Gmail | LinkedIN | GitHub | Medium

Cross-Cutting Capabilities

AI/ML Model Development

Pipeline Automation

Data Integration

Knowledge Engineering

Predictive Analytics

Technical Skills

Machine Learning Graph Neural Networks Deep Learning LSTM NLP LLM Knowledge Graphs GraphRAG GenAI Real World Data WGS RNA Seq scRNA Seq GWAS Cromwell Nextflow Docker RESTful API

Expertise

Bridging research, therapeutic innovation, and technology development through a multidisciplinary capabilities.

Therapeutic Development

1

Target Identification

Discovering and validating therapeutic targets

GWAS Knowledge Graphs Systems Modeling scRNA-Seq RNA-Seq

→

2

Lead Discovery

Designing potential therapeutic candidates

Antibody Design Molecule generation QSPR Models

→

3

Preclinical

Evaluating safety and efficacy

In Silico KO

Research & Precision Medicine

1

Multi-Omics Analysis

Extracting biological insights from complex data

WGS scRNA Seq RNA Seq Proteomics

→

2

Disease Modeling

Understanding disease mechanisms

Knowledge Graphs Boolean Networks Simulation Biomarker Identification

→

3

Clinical Translation

Bridging research to patient care

Temporal KGs Real World Evidence Decision Support

Bioinformatics Pipeline Development

1

Data Acquisition

Ingesting and standardizing biological data

WGS/RNA Seq/scRNA Seq Data Wrangling

→

2

Analysis Pipeline

Building robust analysis workflows

Nextflow Docker Cromwell

→

3

Knowledge Extraction

Deriving actionable insights

ML Models Biomarker Identification Visualization

Translational AI/ML Applications

1

Data Preparation

Structuring biomedical data for AI

Feature Engineering Embeddings Data Augmentation

→

2

Model Development

Building domain specific AI solutions

Applied ML NLP LLM RAG

→

3

Deployment

Implementing solutions in practice

AWS LangChain REST API

Work Experience

AI driven antibody sequence generation

Developed an AI driven pipeline for antibody sequence generation, leveraging deep learning and generative models trained on large scale OAS & SAbDab antibody datasets. Integrated with sequence optimization and structural validation tools to design high affinity, developable antibodies with therapeutic potential.

AntiFold | AbLang | diffAb | BioPhi | LLM fine tuning

Knowledge Graph

Built a KG with open source biological/Real World/Clinical data, harmonized with controlled vocabularies for each entity. Application included drug repurposing, target identification, safety assessment for toxicity and organ wise stratification, reducing months of work to weeks

Neo4j | Link prediction | Node classification | Community Detection

Automated QSPR model building for property prediction

Developed an automated ML pipeline to build Quantitative Structure-Property Relationship (QSPR) model for drug property prediction, that helped reducing dependency on data scientist for model building and increased capabilities across departments.

RDKit | Morgan fingerprint | Optuna - Bayesian Optimization

Machine Learning driven biomarker identification

Developed a computational pipeline for automated biomarker identification using TCGA database, by training classical ML models using RNA seq data, to identify diagnostic and prognostic signatures in BRCA, LIHC, LUAD and PRAD. Enabling faster patient stratification for precision medicine applications.

DEG analysis | WCGNA | Applied ML | SHAP | Kaplan Meier

Network analysis - a systems approach for target identification and prioritization

1. High throughput drug combination (synergy) assessment using network proximity analysis

2. Network vulnerability analysis for target prioritization

Network proximity | Network attributes | STRING

GWAS - WGS

1. GenomeIndia cohort: Performed quality control and variant association analysis followed by cross population studies.

2. AllOfUs cohort: Developed a Cromwell (WDL) pipeline to scale up association studies across 3000+ traits using EHRs to identify statistically significant biomarkers. Association statistics for the study included dosage sensitivity and gene burden test.

Association statistics | Dosage sensitivity | Burden test

In silico KnockOut/perturbation

Developed a high throughput Boolean model simulation pipeline for in silico gene knockout/perturbation experiments, using RNA seq data to initialize the system states supporting data-driven therapeutics which enhances precision in target prioritization.

Attractors | Trap spaces | Logical gates | KEGG

Structure based druggability prediction

Developed a structure based druggability prediction pipeline leveraging parallel processing to accelerate searches across a database of known binding pockets, enabling rapid identification of similar sites to assess target protein druggability.

Graph Analytics | Structural bioinformatics | Multiprocessing

Data driven spectral deconvolution and compound detection

Developed an automated compound identification pipeline leveraging spectral data, matched against a processed in-house reference compound database.

Spectral data processing | Automation

Personal Projects

CAMDA challenge

Constructed a Temporal Knowledge Graph from diabetes patient records using Neo4j, enabling disease progression analysis (trajectory analysis), complication pathways, and early risk prediction
Integrating llama3 with TKG to setup a GraphRAG workflow for various AI driven medical applications

View on GitHub | Read on Medium

Biomedical research assistant

Developed a biomedical research assistant that streamlines literature exploration using AI. The tool retrieves and summarizes top PubMed papers(using PubMed API) with a RAG based pipeline powered by LLaMA 3 and ChromaDB, and includes a chatbot for natural language Q&A. Runs entirely locally using Ollama, ensuring privacy and full control.

View on GitHub | Read on Medium

NeurIPS - Polymer Prediction

A Machine Learning Approach to Polymer Property Prediction: Interpretable Models. Predictive models for Tc, Tg, Rg, FFV and density of polymers.

View on GitHub | Read on Medium

Drug induced liver injury (DILI) prediction model (QSPR)

Build a QSPR model to predict DILI from SMILE, generating descriptors and fingerprints using RDKit and building the model and elucidating mechanistic insights based on SHAP score.

View on GitHub

Protein Classification Models

Built a Deep learning model (using PyTorch) using protein sequence embeddings for protein classification
Developing a Graph Attention Networks (GAT) model for protein structure classification. (work in progress)

View on GitHub

IDA-BRCA

Integrative data analysis (IDA) of Breast cancer (BRCA) dataset for predictive model.

A multi-output classification model using iTRAQ proteome profiles of 77 cancer samples from TCGA.
Classification models using the dataset from METABRIC containing mRNA levels z-score for predicting breast cancer type, tumor stage, and overall survival.

View on GitHub

Natural Language processing

Autism Spectrum Disorder (ASD) classification from caregiver-written behavioral text
Sentiment analysis: Mental health classification
Disaster tweets classification

View on GitHub

Kaggle-projects

Chronic kidney disease prediction using electronic health records.
Pancreatic cancer prediction using urinary biomarkers from urine samples.
Identifed protein biomarkers that can discriminant between different experimental classes of mice with Down syndrome.
Built a plant health prediction model using biosensor data, deployed in Streamlit, and connected it with a REST API for real-time prediction.

View on GitHub