PathGennie enables unbiased simulation of rare molecular events, boosting drug discovery

In the pharmaceutical discovery process, understanding a drug’s residence time—the duration a molecule remains bound to its protein target—is often more important than binding affinity.

A team of scientists has developed PathGennie, a cutting-edge computational framework that dramatically accelerates the simulation of rare molecular events—one of the most challenging problems in computational chemistry and computer-aided drug discovery (CADD). The breakthrough, published in the Journal of Chemical Theory and Computation, offers a powerful and open-source alternative to traditional simulation methods, enabling researchers to predict how drug molecules unbind from proteins without artificial distortions that compromise accuracy.

Overcoming the Bottleneck of Rare Event Simulations

In the pharmaceutical discovery process, understanding a drug’s residence time—the duration a molecule remains bound to its protein target—is often more important than binding affinity. Residence time strongly influences drug efficacy, dosing frequency, and therapeutic potential.

However, simulating the unbinding process is notoriously difficult. These molecular “rare events” occur over milliseconds to seconds, far beyond the practical reach of conventional molecular dynamics (MD) simulations. Even the world's fastest supercomputers struggle to capture such long time-scale events without resorting to shortcuts.

Traditional approaches force unbinding by applying biasing forces or raising the temperature to accelerate the system, but these methods can distort the underlying physics, producing artificially altered pathways that may not reflect real-world behaviour.

PathGennie’s Breakthrough: Direction-Guided Adaptive Sampling

Scientists at the S. N. Bose National Centre for Basic Sciences, Kolkata—an autonomous institute under the Department of Science and Technology (DST)—have introduced a novel approach inspired not by forced dynamics, but by natural selection at the microscopic level.

PathGennie operates by launching swarms of ultrashort, unbiased MD trajectories, each only a few femtoseconds long. It then evaluates these trajectories to determine which ones show genuine progress toward a pre-defined molecular outcome.

• Productive trajectories are extended and evolved,

• Unproductive ones are discarded,

• The system iteratively “selects” the most promising routes.

This “survival of the fittest” strategy allows PathGennie to uncover transition pathways without biasing forces or elevated temperatures. The algorithm operates in any set of collective variables (CVs)—including machine-learned features—giving it unmatched flexibility. It intelligently balances exploration of the molecular landscape with exploitation of promising directions, enabling rapid discovery of realistic kinetic pathways.

Demonstrated Success Across Complex Molecular Systems

In multiple proof-of-concept studies led by Prof. Suman Chakrabarty, along with Dibyendu Maity and Shaheerah Shahid, PathGennie has shown exceptional performance in mapping complex molecular transitions:

• It accurately predicted multiple exit pathways through which benzene escapes from the deep pocket of T4 lysozyme, revealing a network of ligand egress routes.

• It identified three distinct unbinding pathways for the cancer drug imatinib (Gleevec) from Abl kinase, matching previously reported routes that required biased simulations or experimental validation.

These results were achieved without steering forces, yet closely aligned with known experimental and computational findings—offering strong evidence for the method’s accuracy and reliability.

A Versatile Tool for the Global Scientific Community

Beyond ligand–protein interactions, PathGennie’s general-purpose design makes it applicable to a broad range of rare-event scenarios, including:

• chemical reaction mechanisms

• catalytic transformations

• phase transitions

• nucleation and self-assembly processes

Its compatibility with machine learning, especially in defining CVs, opens the door to hybrid physics–AI simulation pipelines that can tackle increasingly complex molecular systems.

With PathGennie made freely available as open-source software, the researchers have lowered the barrier to adopting this technique across academic and industrial settings. The tool promises to significantly accelerate drug discovery workflows, deepen mechanistic understanding in chemistry, and enable physicochemical insights previously out of reach.