A molecular evolution algorithm for ligand design in DOCK

Abstract

As a complement to virtual screening, de novo design of small molecules is an alternative approach for identifying potential drug candidates. Here, we present a new 3D genetic algorithm to evolve molecules through breeding, mutation, fitness pressure, and selection. The method, termed DOCK_GA, builds upon and leverages powerful sampling, scoring, and searching routines previously implemented into DOCK6. Three primary experiments were used during development: Single‐molecule evolution evaluated three selection methods (elitism, tournament, and roulette), in four clinically relevant systems, in terms of mutation type and crossover success, chemical properties, ensemble diversity, and fitness convergence, among others. Large scale benchmarking assessed performance across 651 different protein‐ligand systems. Ensemble‐based evolution demonstrated using multiple inhibitors simultaneously to seed growth in a SARS‐CoV‐2 target. Key takeaways include: (1) The algorithm is robust as demonstrated by the successful evolution of molecules across a large diverse dataset. (2) Users have flexibility with regards to parent input, selection method, fitness function, and molecular descriptors. (3) The program is straightforward to run and only requires a single executable and input file at run‐time. (4) The elitism selection method yields more tightly clustered molecules in terms of 2D/3D similarity, with more favorable fitness, followed by tournament and roulette.

Descriptor-Driven de Novo Design Algorithms for DOCK6 Using RDKit

“…As noted previously, , chemical searching in DOCK_DN can lead to molecules with identical topology but different conformations and/or binding poses. In the present work, to simplify interpretation of D3N outcomes, “duplicate” molecules were removed by (1) grouping all molecules for a given experiment into a single MOL2 file, (2) clustering the molecules based on topological identity using SMILES strings, and (3) retaining only those molecules with the best score depending on the experiment (internal energy, SGE, or MGE).…”

“…The maximum number of scaffold fragments that could be added to any molecule at each layer of growth (dn_max_scaffolds_per_layer) was capped at 1 and the maximum number unsatisfied attachment points per molecule (dn_max_current_aps) at any point was set to 5, which helps control branching. 20 As noted previously, 20,26 chemical searching in DOCK_DN can lead to molecules with identical topology but different conformations and/or binding poses. In the present work, to simplify interpretation of D3N outcomes, "duplicate" molecules were removed by (1) grouping all molecules for a given experiment into a single MOL2 file, (2) clustering the molecules based on topological identity using SMILES strings, and (3) retaining only those molecules with the best score depending on the experiment (internal energy, SGE, or MGE).…”

“…The DOCK6 − program is written primarily in C++ and the most recent version (DOCK6.10) has three main engines for ligand sampling and chemical searching: (1) virtual screening using an anchor-and-grow algorithm, (2) from-scratch construction using a DN design algorithm (DOCK_DN), and (3) molecular evolution using a genetic algorithm (DOCK_GA) . The primary objective of the present work was development and testing of a DOCK6/RDKit interface to allow cheminformatics descriptors to be used in conjunction with DOCK_DN .…”

“…Readers should note that current D3N protocols allow 7 of the aforementioned 13 descriptors to be used during on-the-fly growth. Future work will evaluate the use of other descriptors for D3N as well as develop an RDKit interface for the DOCK6 genetic algorithm (DOCK_GA) recently reported by Prentis et al …”

“…Figure a visually illustrates how fragment libraries are constructed using DOCK6 starting from a collection of input molecules, in this example epinephrine and DANA. Figure b shows 3D representations for 19 fragments, ordered by frequency of occurrence, derived from deconstructing 13,195,579 drug-like molecules downloaded from ZINC (termed the ZINC13M data set). − As in prior work, , for tractability, we choose to retain molecular fragments only if they appeared 13,000 times or more (roughly ∼0.1% of the total), resulting in a final curated set of 382 fragments and 10,844 allowable torsions (bond types). For consistency with prior work, all allowable torsion types were retained.…”