Predicting fragment binding modes using customized Lennard-Jones potentials in short molecular dynamics simulations

Reliable in silico prediction of fragment binding modes remains a challenge in current drug design research. Due to their small size and generally low binding affinity, fragments can potentially interact with their target proteins in different ways. In the current study, we propose a workflow aimed at predicting favorable fragment binding sites and binding poses through multiple short molecular dynamics simulations. Tailored Lennard-Jones potentials enable the simulation of systems with high concentrations of identical fragment molecules surrounding their respective target proteins. In the present study, descriptors and binding free energy calculations were implemented to filter out the desired fragment position. The proposed method was tested for its performance using four epigenetic target proteins and their respective fragment binders and showed high accuracy in identifying the binding sites as well as predicting the native binding modes. The approach presented here represents an alternative method for the prediction of fragment binding modes and may be useful in fragment-based drug discovery when the corresponding experimental structural data are limited.