PathGennie framework accelerates simulation of rare molecular events in drug discovery

PathGennie, a computational framework developed by scientists, can significantly accelerate the simulation of rare molecular events.

Published in the Journal of Chemical Theory and Computation, the open-source software provides a tool for computer-aided drug discovery (CADD) by predicting how potential drugs unbind from their protein targets without the artificial distortions common in standard methods.

In the development of new pharmaceuticals, understanding the “residence time”—how long a drug molecule remains attached to its target protein—is often more critical than binding affinity alone. Simulating the unbinding process, where a drug leaves the protein pocket, is computationally expensive. These “rare events” occur on time scales of milliseconds to seconds, which are challenging or impossible to access using standard classical molecular dynamics (MD) simulations, even with powerful supercomputers.

Traditionally, scientists force these events to occur by applying artificial bias forces or elevated temperatures, which can distort the physics of the interaction and lead to inaccurate predictions of the transition pathways.

Researchers at S. N. Bose National Centre for Basic Sciences, Kolkata, an autonomous institute of the Department of Science and Technology (DST), developed PathGennie to mimic natural selection on a microscopic scale instead of forcing the molecule to move.

The algorithm launches swarms of ultrashort, unbiased molecular dynamics trajectories, each only a few femtoseconds long, and selectively extends those trajectories that make progress toward a desired outcome.

In essence, it acts as a direction-guided “scouting” mission in the molecule’s conformational landscape. Numerous tiny simulation snippets are initiated, and those moving closer to a defined end state are prolonged, while unproductive ones are discarded. This “survival of the fittest” approach allows the algorithm to bypass long waiting times of rare events without applying external biases or elevated temperatures, retaining the true kinetic pathways.

The method is general and can operate in any set of collective variables (CVs), including high-dimensional or machine-learned CV spaces. By dynamically balancing exploration and exploitation, PathGennie quickly identifies transition pathways that would otherwise require prohibitively long simulations to discover.

In proof-of-concept studies, PathGennie, created by a team led by Prof. Suman Chakrabarty along with Dibyendu Maity and Shaheerah Shahid, demonstrated the ability to uncover multiple competing pathways for several molecular systems.

For example, it rapidly mapped how a benzene molecule escapes from the deep binding pocket of the T4 lysozyme enzyme, revealing a network of distinct ligand exit routes. The algorithm also identified three separate dissociation pathways for the anti-cancer drug imatinib (Gleevec) as it unbinds from the Abl kinase. The pathways matched all previously reported routes in the literature with just a few iterations. The ligand unbinding pathways were found without steering forces and matched mechanisms seen in earlier biased simulations and experiments, validating PathGennie’s accuracy.

Because PathGennie is a general-purpose framework, it can be adapted to a wide range of rare events beyond those tested. The authors note that it is immediately applicable to problems such as chemical reactions, catalytic processes, phase transitions, or self-assembly phenomena, where one needs to find a transition pathway over a high energy barrier.

The software is also compatible with modern machine-learning techniques, allowing machine-learned order parameters to guide the sampling. PathGennie has been made freely available to the scientific community, lowering the barrier for other researchers to use the technique.