Integrated Approach of Machine Learning and High-Throughput Screening to Identify Chemical Probe Candidates Targeting Aldehyde Dehydrogenases

Abstract

Selective chemical probes are essential for dissecting biological pathways and advancing drug discovery, yet developing high-quality probes for targets such as the aldehyde dehydrogenase (ALDH) family remains challenging. Here, we present a novel integrated approach combining experimental quantitative high-throughput screening (qHTS) with advanced machine learning (ML) and pharmacophore (PH4) modeling to rapidly identify selective inhibitors across multiple ALDH isoforms. We screened ∼13,000 annotated compounds against biochemical and cellular assays. We then utilized the data set to build ML and PH4 models to virtually screen a larger set of 174,000 compounds to enhance the chemical diversity of hits. This approach led to the expansion of chemically diverse isoform-selective inhibitors that are potent in both biochemical and cell-based assays. Validation through cellular target engagement assays further confirmed the selective activity of these compounds, leading to the discovery of ALDH1A2, ALDH1A3, ALDH2, and ALDH3A1 chemical probe candidates. Remarkably, this was achieved by employing just a single iteration of quantitative structure–activity relationship (QSAR) and PH4 modeling for virtual screening. This combined in vitro and in silico strategy not only enhances the discovery of biologically relevant chemical probe candidates but also significantly expands the chemical diversity accessible for probe development, establishing a new platform for the rapid and resource-efficient identification of chemical probes against the ALDH enzyme family. The data set generated, including hundreds of compounds thoroughly characterized across a spectrum of assays, is publicly available and can serve as a high-quality training set for future research initiatives and probe development efforts.