| Savitribai Phule Pune University |
| Title | Advancing Drude polarizable force field to drug like molecules using neural network design |
| Speaker |
Anmol Kumar,
School of Pharmacy, University of Maryland, Baltimore, USA
Dr Anmol Kumar obtained his PhD in Chemistry from IIT Kanpur in 2017. As a PhD student, he contributed to the analysis of molecular electrostatic potential and its gradient properties. He joined as a postdoctoral research fellow at the School of Pharmacy, University of Maryland Baltimore with Professor Alexander D MacKerell. His work includes developments in the CHARMM package, QM/MM simulations, CHARMM force-field development, and applications of machine learning in force-field. One of his important contributions from his postdoctoral laboratory is the FFParam package. He has recently rejoined Professor Alexander's group as research associate. |
| Date & Time | Monday, 23 January 2023 | 15:00-16:00 |
| Venue | PC RAY HALL, Department of Chemistry, SPPU, Pune |
| Abstract |
CHARMM Drude force-field (FF) is a well-established atomistic FF for biomolecules such as proteins, nucleic acids, lipids, and carbohydrates. Its ability to capture electronic polarization effects via auxiliary particles (Drude oscillators) attached to non-hydrogen atoms sets it apart from commonly applied additive FFs, which rely on fixed charges. Extension of Drude FF to novel drug-like molecules is challenging as it requires a mechanism to assign electrostatic (atomic charges as well as polarizabilities and Thole scale factors), van der Waals, and bonded parameters to non-biomolecular systems. These parameters are initially obtained by fitting to quantum mechanical (QM) target data and further validated to reproduce bulk-phase thermophysical properties.
We developed a deep neural network (DNN) model that can determine RESP charges on atoms as well as lone pairs (an integral feature of Drude FF) and atomic polarizabilities (Alpha) and Thole scaling factors (on non-hydrogen atoms) without the necessity of performing QM calculations. This model was trained on charges and polarizabilities obtained from high level QM calculations of a very large molecular training dataset. In conjunction with electrostatic parameters, we have also put significant effort towards the determination of Lennard-Jones (LJ) parameters using a set of DNN models for various atom types belonging to conjugated alkenes, alkynes, nitriles, amines, nitro-benzyl species, bipyrroles and so on. The current bonded parameter assignment is based on analogy with existing Drude FF parameters of biomolecules. Efforts are underway to build a model for extending the scope of bonded-parameter prediction which utilizes global parameter optimization protocol targeting QM optimized geometries and potential energy scans (PES). Our multifaceted approach towards optimizing all components of Drude FF using large amount of chemical data will harbinger the extension of Drude FF to novel drug-like molecules. The details of Drude FF, its applications and our approach towards its extension will be discussed in detail. |
| Organizer/Host | SCMS-SPPU |
| Slides/Poster |
Wish to subscribe to our announcements list? Send blank email with subject "subscribe" to announce-join AT scms.unipune.ac.in