Harnessing Computational Chemistry and Structural Biology Platforms for in silico Drug Discovery

Abstract

The integration of computational methods in drug discovery has become essential, necessitating the development of open-source, modular, and reproducible workflows that are adaptable to an evolving field. In this context, the synergistic combination of molecular mechanics (MM) and machine learning (ML) offers avenues for expediting the identification and optimisation of potential drug candidates in the hit-to-lead stage. Automated free energy calculations for the prediction of binding free energies of congeneric series of ligands to a protein target are growing in popularity, but choosing transformations and building reliable initial binding poses for the ligands remains challenging. In this thesis, an open-source python package, FEgrow, is presented. This package automates the required construction and evaluation of congeneric compound series within protein binding pockets, by employing hybrid ML/MM potential energy functions. FEgrow optimises suggested compounds’ bioactive conformers using physics-based methods and scores them using a convolutional neural network (CNN) scoring function, rapidly finding relevant areas of chemical space, as well as generating accurate 3D structures which can then be utilised in more rigorous calculations, such as free energy perturbation (FEP). This workflow was applied to the CACHE#2 (Critical Assessment of Computational Hit-finding Experiments) Challenge, which serves as a validation exercise that aims to establish benchmarks in molecular design by providing high-quality experimental feedback on in silico design predictions. In this challenge, compounds designed via FEgrow yielded multiple low micromolar hits for the NSP13 helicase of SARS-CoV-2, demonstrating the efficacy of the workflow. To further streamline the FEgrow workflow, an updated Active Learning (AL) approach along with the utilisation of large on-demand library searches (such as Enamine REAL) was developed. This addition enhances the exploration of chemical space for de novo design and was validated through experimental assays, where three designed compounds showed weak activity in a fluorescence-based Mpro assay.