At Recursion, we are on a mission to radically improve patient lives by revolutionizing drug discovery through technological innovation and multi-modal insight generation. Here we describe three preclinical pipeline platform methods that can be assembled in a modular fashion to identify and achieve key stage gate milestones in hit-to-lead and lead-optimization while simultaneously reducing timelines, cutting costs and increasing scale. We show examples from our preclinical pipeline demonstrating the utility of these methods to drive target agnostic discovery of novel first-in-class small molecule oncology therapeutics.(1) Early MoA prediction and derisking: Utilizing our powerful phenomics platform, we can identify starting novel chemical matter predicted to impact disease-relevant biology. However, within most disease-relevant areas of biology, there are strategic or toxicity-based rationales to avoid certain MoAs. By using a combination of phenomics and structural insights, we have built a powerful MoA prediction tool that provides an early view into compound mechanistic insights at the initial hit stage. Hit compounds can then be tested in biochemical assays informed by the predictive model to validate desirable MoAs and/or de-risk undesirable MoAs and therefore advance only those with a desirable path forward. (2) Translational model selection: The pursuit of first-in-class approaches combined with target agnostic discovery presents a challenge to early discovery teams seeking to select best-fit translational models, as the information and predictive value provided by literature is limited. Here we show an example of how we can leverage a combination of public large-data insights, Tempus patient data, DepMap dependency scores, and phenomic relationships to select preclinical translational models that are well suited for testing our hypotheses. Such selection strategies can be automated, enabling rapid execution and decision making in the hit-to-lead stage of discovery. (3) Multimodal SAR: We leverage both phenomics and transcriptomics to drive structure-activity relationship (SAR) optimization in a target agnostic fashion. We integrate these two data layers to select novel chemical entities (NCEs) that drive our biology of interest in both orthogonal assay modalities, enabling us to select compounds with the best in vivo potential. We show an example of how to drive lead optimization SAR by comparing the phenomic and transcriptomic signature of NCEs to those of an efficacious hit compound that potentiates immunotherapy response in vivo, resulting in a polypharmacologic profile that eradicates tumors in 100% of treated animals. Because our approach does not require target-based bespoke cellular or biochemical based potency assays, SAR is highly scalable and is currently being utilized in other programs with a diversity of MoAs.

VIEW THE PUBLICATION