Aaryesh Deshpande

About Me

Graduate Research Assistant • Machine Learning & Computational Biology

Georgia Institute of Technology

I'm an M.S. Bioinformatics student at Georgia Tech working in the field of machine learning and computational biology. My research focuses on modeling protein dynamics and function, leveraging protein language models, graph neural networks, and physics-aware learning systems.

I'm interested in designing practical computational tools and models that bridge machine learning with protein dynamics. Outside research, you'll usually find me experimenting with filmmaking, animation, and motion design, because storytelling, whether in science or cinema, is half the craft.

Based in Atlanta, GA

Open to collaborations & research opportunities

Projects

Transformer-Based Generation of Novel Drug-like Molecules

July - Nov 2023

A fine-tuned Transformer model mapping gene expression perturbations to molecular SMILES for targeted cancer drug design; achieved 23% novel scaffold generation with high pathway alignment and QED >0.75.

Transformers GEO Drug Design Generative Model

GU Drug-Pro Toolkit for Drug Property Prediction

Dec 2023 - May 2024

A full-stack ML-based bioinformatics toolkit integrating 48 optimized models for ADMET prediction, lead selection, and predictive analytics in early-stage drug discovery.

Machine Learning ADMET Drug Discovery

Sequence Clustering & Consensus Generation Tool

Jan - Feb 2025

A multi-stage clustering algorithm combining sequence embeddings and conformational similarity metrics for ensemble reduction and consensus selection in MD workflows.

Clustering MD Simulation Selection algorithm Embeddings

Microprotein Discovery Bioinformatics Toolkit (smorfari)

March - Oct 2025 Active Testing

An open-source end-to-end bioinformatics tool for microprotein (smORF) discovery and annotation, integrating ORF prediction, sequence similarity, and annotation networks with taxonomic and domain-aware filters.

Bioinformatics smORF Annotation Genomics

Molecular Graph Encoder with Dynamics Reconstruction

April - July 2025

Designed a protein residue-level graph encoder-decoder framework integrating SE (3)-equivariant molecular graph embeddings with a latent dynamics’ reconstruction module. The model enforces Markovian evolution in latent space for CV identification and reconstructs free-energy landscapes and trajectory-consistent conformations using physics-informed constraints, aiming for transferable recovery of dynamics across homologs.

Graph Neural Networks Encoder-Decoder Free Energy PIML

State-Conditioned Fusion Hypergraph for PPARγ Ligand Toxicity

September 2025 - Present Ongoing

A dual-encoder DTI model with state-conditioned hypergraph fusion. The system encodes ligands via pharmacophore-aware hypergraphs and protein pockets via state-specific residue graphs (agonist vs antagonist), fusing representations through cross-attention for ligand toxicity and functional state classification.

Hypergraphs DTI Model Cross-Attention Toxicity Prediction Pharmacophore

Learning Biomolecular Motion: The Physics-Informed Machine Learning Paradigm

arXiv:2511.06585 • November 2024

A comprehensive review of physics-informed machine learning approaches for modeling and predicting biomolecular dynamics, bridging traditional molecular simulations with modern deep learning techniques.

arXiv

Review Paper Physics-Informed ML Biomolecular Dynamics

Hi! 👋🏼 I'm Aaryesh Deshpande

About Me