Pharmacophore-based design by learning on voxel grids

TL;DR

The paper addresses the scalability bottlenecks of pharmacophore-based virtual screening by introducing VoxCap, a voxel-based pharmacophore-to-SMILES generator. It presents two practical workflows—de-novo generation and a fast search pipeline—that jointly enable generation of diverse, 3D-consistent hits and efficient hit identification in large libraries. VoxCap, evaluated against the PGMG baseline on GEOM-drugs and ChEMBL, yields markedly more hits and unique scaffolds, while the fast search dramatically reduces 3D comparison costs, albeit with some loss in total hits. These findings advance scalable, structure-aware ligand design and enable practical exploration of vast chemical spaces for drug discovery.

Abstract

Ligand-based drug discovery (LBDD) relies on making use of known binders to a protein target to find structurally diverse molecules similarly likely to bind. This process typically involves a brute force search of the known binder (query) against a molecular library using some metric of molecular similarity. One popular approach overlays the pharmacophore-shape profile of the known binder to 3D conformations enumerated for each of the library molecules, computes overlaps, and picks a set of diverse library molecules with high overlaps. While this virtual screening workflow has had considerable success in hit diversification, scaffold hopping, and patent busting, it scales poorly with library sizes and restricts candidate generation to existing library compounds. Leveraging recent advances in voxel-based generative modelling, we propose a pharmacophore-based generative model and workflows that address the scaling and fecundity issues of conventional pharmacophore-based virtual screening. We introduce \emph{VoxCap}, a voxel captioning method for generating SMILES strings from voxelised molecular representations. We propose two workflows as practical use cases as well as benchmarks for pharmacophore-based generation: \emph{de-novo} design, in which we aim to generate new molecules with high pharmacophore-shape similarities to query molecules, and fast search, which aims to combine generative design with a cheap 2D substructure similarity search for efficient hit identification. Our results show that VoxCap significantly outperforms previous methods in generating diverse \textit{de-novo} hits. When combined with our fast search workflow, VoxCap reduces computational time by orders of magnitude while returning hits for all query molecules, enabling the search of large libraries that are intractable to search by brute force.

Figures (3)

• Figure 1: Example molecule from the ChEMBL dataset with its molecular graph, 3D shape profile and 3D pharmacophore profile. Each colour in the pharmacophore profile (one for each pharmacophore type as shown in the legend) is represented as a different channel to the model, as is the shape profile. Distances on the axes are measured in Angstroms. For visualisation purposes, we threshold the voxel grids displayed here such that values under 0.1 are set to 0.
• Figure 1: Model architecture
• Figure 2: 3D vs 2D similarities, from fast database search experiment with $n_g=500,~n_a=1$