I'm a PhD researcher (Senior Research Fellow) in the Department of Chemical and Biological Sciences, S. N. Bose National Centre for Basic Sciences, Kolkata. I use molecular simulation and data-driven methods to find allosteric and cryptic binding sites and to modulate protein–protein interactions (PPIs) for drug discovery.
- 🔬 Current focus: mixed-solvent / mixed amino-acid MD for PPI hotspot and cryptic-pocket mapping (PLK1 PBD, PCSK9)
- ⚗️ Methods: all-atom MD, enhanced sampling, alchemical free energy, neural-network potentials (ANI-2x in GROMACS), docking and ML rescoring, structural-ensemble generation (BioEmu)
- 🎯 Looking for: postdoctoral positions in computational biophysics, allostery and structure-based drug design
Cosolvent probes sample the protein surface, residing longer at PPI hotspots and seeding a transient cryptic pocket, the idea behind PPIscout and my PLK1 / PCSK9 allosteric-pocket work.
-444d56?style=flat-square){kind=icon}
|
De novo design against the TEM-1 β-lactamase cryptic allosteric pocket, with three generative paradigms (CReM, SAFE-GPT, SELFIES-GA) and a docking protocol that has to pass redocking before it is used. The site-selectivity claim is tested against the two ways it could be an artefact — selection bias (which inflated it 68%) and unequal box volumes (which, tested, had been understating it) — leaving a +1.3 kcal/mol paired preference over the catalytic site.
|
|
Can ligand structure alone tell an allosteric modulator from an orthosteric one? A ChEMBL 37 benchmark whose point is how much apparent accuracy is artefact: ROC-AUC falls from 0.995 (random split) to 0.723 (held-out target), and a probe that sees only which protein a compound was tested on — no chemistry at all — still scores 0.950 on a random split. It also shows that pooled physicochemical comparisons invert the within-target ones (a Simpson's paradox).
|
|
Interaction-fingerprint (PLIF) rescoring of AutoDock Vina poses for early enrichment in virtual screening. It uses scaffold-aware cross-validation, LightGBM on ProLIF bitvectors, and BEDROC/EF with bootstrap confidence intervals. It removes Vina's ligand-size bias and recovers the canonical CDK2 hinge Leu83 H-bond as the top discriminative feature.
|
| Year | Publication |
|---|---|
| 2026 | Mapping PPI hotspots and unveiling a cryptic allosteric pocket in PLK1 PBD via mixed-solvent MD — ChemPhysChem · doi |
| 2026 | Computational strategies for allosteric drug discovery: from cryptic pocket detection to rational design — Chem. Commun. · doi |
| 2025 | Mixed-solvent MD reveals a druggable allosteric pocket in the PCSK9 C-terminal domain — J. Phys. Chem. B · doi |
| 2025 | PPIscout: PPI hotspot mapping using mixed amino acid–water MD — J. Chem. Sci. · doi |
| 2025 | Harnessing allostery to modulate protein–protein interactions: from function to therapeutic innovations — J. Mol. Biol. · doi |
| 2024 | Conformational and binding behaviour of human serum albumin induced by surface-active ionic liquids — J. Phys. Chem. B · doi |
| 2024 | Allosteric hotspots and cryptic sites to modulate PPIs: a molecular thermodynamic approach — Biophys. J. · doi |
| 2022 | A multidrug efflux protein in M. tuberculosis: Tap as a drug-repurposing target — Comput. Biol. Med. · doi |
Full list on Google Scholar.
