A Multi-Omics Interpretable Machine Learning Model Reveals Modes of Action of Small Molecules (ChIP-Seq)

High-throughput screening and gene signature analyses frequently identify lead therapeutic compounds with unknown modes of action (MoAs), and the resulting uncertainties can lead to the failure of clinical trials. We developed a multi-omics approach for uncovering MoAs through an interpretable machine learning model of the effects of compounds on transcriptomic, epigenomic, metabolomic, and proteomic data. We applied this approach to examine compounds with beneficial effects in models of Huntington’s disease, finding common MoAs for previously unrelated compounds that were not predicted based on similarities in the compounds’ structures, connectivity scores, or binding targets. We experimentally validated two such disease-relevant MoAs, autophagy activation and bioenergetics manipulation. This interpretable machine learning approach can be used to find and evaluate MoAs in future drug development efforts. Cells expressing mutant huntingtin were treated in triplicate with serum-free DMEM with vehicle (Q111SST) or serum-free DMEM with one of 14 protective compounds for 24 hours. Wild type cells were also treated with serum-free DMEM with vehicle (Q7SST) as an additional control for 24 hours. We examined the compounds' epigenomic effects on the cells using H3K4me3 ChIP-Seq.