Portfolio
Personal statement, appointments and honors, and contributions to science.
I am a computational biophysicist with 10+ years of experience developing and applying molecular modeling tools to investigate fundamental biomolecular processes relevant to medicinal chemistry and biochemistry. During my PhD at the Indian Association for the Cultivation of Science, I developed quantum-mechanics/molecular-mechanics (QM/MM) methods1 and machine-learning-based force fields for molecular dynamics (MD) simulations aimed at understanding photophysical processes in biology2. In my postdoctoral training in Dr. Alex Dickson's laboratory at Michigan State University, I gained expertise in enhanced-sampling MD methods — essential tools for understanding long-timescale, rare events in biology (e.g., drug binding, protein folding, large-scale conformational switches, and molecular transport through transmembrane proteins). I developed a robust enhanced-sampling technique3 by integrating the weighted-ensemble (WE) MD method with Markov state models (MSMs) to decipher the pathways of slow, pharmacologically relevant inhibitor (un)binding in soluble epoxide hydrolase, and provided key insights for increasing ligand residence time4 — a crucial determinant of pharmacokinetic efficacy. I have also applied WE simulations to explore the long-timescale membrane-association dynamics of the PIKfyve kinase, otherwise inaccessible to conventional MD.
My long-term objective is to establish an independent research lab that leverages state-of-the-art computational techniques to advance our understanding of pharmacological and biochemical mechanisms. To this end, I have recently begun an independent position as a Fixed-Term Assistant Professor, where I am committed to fostering interdisciplinary collaborations that integrate molecular simulation with experimental approaches in structural and molecular biology. I have collaborated with several experimentalists and applied methods such as WE, MSM, and unbiased MD to study molecular mechanisms, thermodynamics, and kinetics across a wide range of biomolecular systems. Most recently, I have been working with Dr. Benjamin Orlando to investigate substrate-transport pathways in the transmembrane protein CbrA from Pseudomonas species using long, unbiased MD simulations. I plan to extend this work by applying enhanced-sampling techniques to probe the energetics and mechanisms of molecular transport and signaling in CbrA. By integrating structural and biochemical results with molecular simulation, our ultimate goal is to provide a comprehensive understanding of the principles that allow CbrA to act as a transceptor central to Pseudomonas virulence and pathogenesis.
2024 – present Fixed-Term Assistant Professor, Michigan State University, East Lansing, MI
2021 – 2024 Postdoctoral Research Associate, Michigan State University, East Lansing, MI
2017 – 2020 CSIR Senior Research Fellow & PhD fellow, Indian Association for the Cultivation of Science, Kolkata, India
2014 – 2017 CSIR Junior Research Fellow, National Chemical Laboratory, Pune, India
2017 Physical and Materials Chemistry Division Day poster award, National Chemical Laboratory, India
2014 National Eligibility Test (NET) PhD fellowship in Chemistry, Council of Scientific and Industrial Research, India
2014 Graduate Aptitude Test in Engineering (GATE) PhD fellowship, National Coordination Board — GATE, India
Modeling rare but biologically significant events — such as ligand unbinding or long-timescale protein conformational changes — is essential for understanding the fundamentals of biochemistry. Traditional MD simulations cannot capture these long-timescale processes because of their computational cost. My research has contributed to the development and application of enhanced-sampling techniques that overcome this limitation, enabling detailed study of rare events with pharmacological relevance. I have used WE-based methods to investigate the unbinding mechanisms of several inhibitors targeting soluble epoxide hydrolase (sEH)a. These inhibitors span a wide range of experimentally measured residence times, a pharmacokinetic property directly related to drug efficacy. By capturing statistically meaningful ensembles of unbinding pathways, we characterized the transition state ensembles (TSEs) for each ligand, revealing that structurally similar bound states can give rise to surprisingly diverse TSEs. These results underscore the complexity of computational drug design and highlight the importance of simulating transition states directly, rather than relying solely on static structures or thermodynamic endpoints. We also show that, although standalone WE methods provide reasonable mechanistic insight, they are not quantitatively accurate in describing the underlying slow kinetics that are crucial in kinetics-driven drug discovery. In a subsequent study, we propose that combining MSMs with WE methods increases the statistical significance of kinetic predictions. This is not straightforward, however; we identify the methodological challenges of integrating MSMs with WE data and build an analytical framework for constructing WE-MSMs. Applied to the long-timescale unbinding of inhibitors from sEH, our method achieves a significant improvement (more than an order of magnitude) in accuracy and robustness compared with standalone WE or MSM methodsb. Collectively, these contributions advance computational methods for studying rare, long-timescale events in biologically and pharmacologically important systems — enhancing both mechanistic understanding and predictive power for the rational design of more effective therapeutics.
Fluorescent proteins and their synthetic analogs are indispensable tools in biomedical imaging and molecular diagnostics. A quantitative understanding of how the local environment modulates the spectral properties of these chromophores is essential for rational probe design. My research has used polarizable hybrid QM/MM methods to dissect and predict environment-induced spectral shifts in chromophores, particularly those derived from green fluorescent protein (GFP). Using a polarizable QM/MM framework based on the effective-fragment-potential (EFP) method, I investigated the solvatochromic behavior of a GFP chromophore variant in watera. While vertical excitation energies (VEEs) showed minimal solvent-induced shifts (~0.1 eV), the vertical detachment energies (VDEs) showed a pronounced red shift, validating a long-standing hypothesis about significant charge reorganization during ionization. In a subsequent study, I integrated QM/MM with MD to simulate and characterize the absorption spectra of the GFP chromophore variant in the RNA Spinach aptamerb. Our hybrid method reproduced the red shift observed in the complex RNA Spinach environment. Contrary to prior assumptions that π-stacking interactions between chromophore and nucleobases drove the shift, our simulations showed that long-range electrostatics between the chromophore and the negatively charged RNA scaffold played the dominant role. We further uncovered an interplay between opposing electrostatic influences: the RNA backbone and surrounding polarized water molecules exerted partially cancelling effects that together shaped the observed shift. These findings emphasize the complex heterogeneity of biomolecular environments and their nontrivial influence on chromophore photophysics, and establish the utility of polarizable QM/MM models for interpreting and predicting environment-induced spectral changes — a pathway toward more sensitive, tunable optical probes in complex biological settings.
Accurate prediction of spectral properties such as ionization energies (IEs) and electron affinities (EAs) in complex biological environments requires extensive sampling, because solvation dynamically affects the chromophore structural ensemble. Traditional QM/MM approaches are often prohibitively expensive in this context, requiring thousands of calculations to predict properties such as spectral peak maxima and full widths at half maximum. My research addresses this bottleneck through a solute–solvent interaction-energy-driven descriptor that can serve as a reaction coordinate for the rapid sampling of critical snapshots in a spectral shapea. This biased sampling significantly accelerates the convergence of spectral properties while maintaining high accuracy, without explicit QM/MM calculation at every configuration. The approach has been validated across multiple systems — phenol, thymine, and the GFP chromophore in the condensed phase — with consistent acceleration in predicting ionization energies and spectral widths. In a subsequent collaborative study on a much larger guanine nucleotide in water, a high-level QM/MM framework combined the EFP method (MM) with the equation-of-motion coupled-cluster technique for ionization potentials (EOM-IP-CCSD; QM). The expense of these methods required minimal sampling of critical snapshots; even so, the biased-sampling technique recovered the vertical ionization energies and spectral properties accurately with only a handful of expensive QM/MM calculationsb. Overall, these contributions advance the computational toolkit for simulating the spectral properties of photophysically relevant biomolecules — facilitating broader, more detailed studies of light–matter interactions, an essential component for understanding radiation-induced DNA damage and photochemical reactivity in biology.
Accurate and efficient estimation of interatomic interaction energies is foundational to molecular simulation. Classical force fields offer speed but often sacrifice accuracy through simplified empirical forms, whereas quantum-mechanical (QM) methods provide high precision but are prohibitively expensive for large systems. My research addresses this trade-off by developing machine-learning (ML) force fields that combine QM accuracy with classical efficiency, particularly for water — a fundamental yet notoriously challenging system because of its strong polarization effects. I explored support-vector regression (SVR) as an alternative to neural-network methods; by integrating SVR with the many-body-expansion (MBE) framework, we achieved highly accurate interaction-energy predictions for water clusters. For the water decamer, the method achieved less than 2% ML error relative to the parent QM calculations in absolute interatomic interaction energy and showed transferability across temperaturesa. To validate the model in real-world simulations, we implemented it in Monte Carlo (MC) simulations and compared the resulting ensemble properties with those from ab initio MD (AIMD) and the classical TIP3P model. The ML-based MC (MLMC) simulations reproduced key structural metrics — radial distribution functions, tetrahedral order parameters, and hydrogen-bonding patterns — with remarkable fidelity to AIMD, whereas the classical model showed noticeable deviationsb. Collectively, these studies represent a significant advance toward next-generation force fields: by leveraging ML within a physically grounded framework such as MBE, and deploying it successfully in simulation protocols such as MC, my work demonstrates a pathway to highly accurate, scalable, and computationally tractable modeling of complex molecular systems.