Research Experience

AI Research Scientist | Exscientia, Miami, US

Working with the AI-ML team to expedite drug discovery by developing hybrid AI/QM/MM methods. Leading PhysicsML team (within ML team).

Discovered graph-based neural network potential (EXS-NNP) outperforming state-of-the-art ANI on coupled-cluster theory dataset, recognized as a significant breakthrough at Exscientia.
Enhanced efficiency of NNPs by reimplementing them for multi-GPU training and predictions, automated training workflow, and integrated for conformer generations and MD simulations.
Research project (10% innovation): “Fusing Force-Field Features within GNNs”: Crafted a novel message- passing GNN architecture bridging AI & Physics, predicting fast and accurate energies/binding-affinity.

Bridging the gap between ML teams and chemists by actively engaging with medicinal chemists to refine AI models based on their feedback.

Projects involved: Diffusion Model Based Docking; to mitigate protein-ligand clashes.
Active Learning for Compound Selection: to refine binding free energy predictions,
Implemented a docking workflow to generate 3D structural features, and integrating these features with 2D ADMET models, resulting enhanced overall performance.
Successfully reintroduced Deep Learning based AI-based linker generation between two fragments, resulting in the submission of two molecules for synthesis and in-vitro assays

Implemented generative deep learning model to guide conformational sampling of intrinsically disordered proteins (IDPs) and to reduce dimensionality.
Developed a ML-enabled pipeline for large-scale virtual drug screening using clustering and deep learning in combination with physics-based approaches against RPN11 - a drug target for breast cancer.
Designed an efficient workflow for hit-to-lead drug discovery by integrating a hybrid neural-network/classical potentials molecular dynamics simulations against COVID-19 main protease (MPRO).
Other areas: Modeled electron transport in peptide helices subject to chirality induced spin selectivity (CISS) effect and its application in protein-protein association.

Evaluated performance of deep neural network potentials (ANI) to achieve DFT accuracy at force-field speed (100x) for small molecule crystal polymorph prediction. Reported 98% DFT-ANI correlation in prediction.
Analyzed potential energy surfaces of 100 different crystal structures (over 500 polymorphs) using DFT/ANI-1 potentials and identified their experimental stable forms (from exhaustive literature search).

Performed quantum chemistry calculations
QM techniques used: DFT, TDDFT, QMMM, AIMD, PES, Electron Transfer, Molecular Orbitals, NBO, ESP, NMR, Spectra, QST2, CASSCF.

Moscow Institute of Physics and Technology, Moscow, Russia | Advisor: Prof. Artem Oganov
o Used evolutionary algorithm (USPEX) to predict the stable and metastable structures of europium nitride (with the chemical composition only).
o Reported the most crystal stable structure in reference to the experimental literature.
Indian Institute of Sciences (IISc), Bengaluru | Advisor: Prof. S Yashonath
o Performed nanoscale molecular dynamics simulations of Zeolite MOF using DLPOLY package.
o Analyzed different warfare agents (Xylene, Benzene etc) adsorbed on the Zeolite surface.
Bhabha Atomic and Research Centre, Mumbai | Advisor: Prof. Swapan Ghosh
o Carried out periodic DFT on novel porous carbon nitride (C3N4) as a catalyst for photooxidation of water.
o Investigated 4e- and 2e- water splitting reactions (DFT calculations using GAMESS package).
National Institute of Interdisciplinary Science and Technology (CSIR) | Advisor: Dr. CH Suresh
o Modeled reaction isomerization path (cyclopropene to allene) using first principle methods.
o Proposed a new reaction pathway with smaller activation energy (performed transition state DFT calculations).
Institute of Chemical Technology | Advisor: Prof. N. Sekar
o Predicted color of dye molecules using theoretical calculations - particle in a box/ring methods.
o Theoretical model was evaluated against the experimental references to test its efficiency.

Thesis titles: “Dyeing with Fluorescent Dyes” & “Computational insight into possible dehydrated and depolymerized mechanisms of cellulose”
Advisor: Dr. U Sayyed | Grade : A
Unforeseen bending in 1D silicene layers - Yearlong work in Prof. VG Gaikar group

Work Description: Synthesis workflow of industrial auxiliaries and chemicals (wet-lab experiments)