De novo generation of multi-target compounds using deep generative chemistry

Abstract

Polypharmacology drugs—compounds that inhibit multiple proteins—have many applications but are difficult to design. To address this challenge we have developed POLYGON, an approach to polypharmacology based on generative reinforcement learning. POLYGON embeds chemical space and iteratively samples it to generate new molecular structures; these are rewarded by the predicted ability to inhibit each of two protein targets and by drug-likeness and ease-of-synthesis. In binding data for >100,000 compounds, POLYGON correctly recognizes polypharmacology interactions with 82.5% accuracy. We subsequently generate de-novo compounds targeting ten pairs of proteins with documented co-dependency. Docking analysis indicates that top structures bind their two targets with low free energies and similar 3D orientations to canonical single-protein inhibitors. We synthesize 32 compounds targeting MEK1 and mTOR, with most yielding >50% reduction in each protein activity and in cell viability when dosed at 1–10 μM. These results support the potential of generative modeling for polypharmacology.

Structure-Based Drug Design with a Deep Hierarchical Generative Model

“…Rather than further optimizing the exhaustive search of chemical space, deep generative models are capable of learning distributions over molecular data. These models have already been successfully applied to large drug-like chemical data sets for de novo molecular generation of drug-like molecules − and discovery of new potent hit molecules. − Lately, efforts have turned toward modeling drug-target interactions by training models on data sets of ligand-bound receptor crystal structures. − By learning a joint probability distribution of ligands bound to their receptors, these models can generate new molecules conditioned on the receptor, effectively narrowing the regions of chemical space in which to search for new drugs. Because generative models are generally trained in an unsupervised fashion to accurately reproduce features of the training data set, their applicability will both be defined by and improve with the size, scope, and quality of available data.…”

A Missionary Training Program For University Students In South Korea

“…Although the deep generative models have proved useful in generating novel and chemically valid molecules, further screening is necessary to evaluate their potential to bind with specific protein targets. Building on this notion, several researchers have developed target-specific molecule generation models to produce novel, drug-like molecules that are highly likely to interact with specific target proteins, including Transformer-based generation (Grechishnikova, 2021), AlphaDrug (Qian et al, 2022, SiamFlow (Tan et al, 2022) and POLYGON (Munson et al, 2024).…”