BenevolentAl Therapeutics Pipeline and Triage

Target Prediction Diverse training data yields high-quality Al inferences Literature and Knowledge Graph Training Data [Gene] inhibitor attenuates [Disease] [Gene] regulates [Biological Process] Targeted inhibition of [Gene] limits [Disease] + High-Quality Assay Data Previous Disease Programs CD-8 T Cell Activation Endothelial LDL regulation B Cell CD40 Positive Regulation IL-6 IL-8 IP-10 MCP-1 Algorithms identify sentence forms that suggest strong links by expanding on small curated ground-truth datasets Internal experts define a subset of key relationships that model biomedical knowledge from the scientific literature Our large scientific text corpus results in very high recall across diseases and mechanisms Internal experts curate public and private high-quality transcriptomics datasets Datasets are grounded to diseases, mechanisms, and proteins in the knowledge graph, allowing prediction and evaluation from the knowledge graph representation Prior disease programs provide results for training subsequent models Three kinds of information are routinely captured and available for training: Hit/ no-hit Ranked assay results Triage annotations and reasoning (safety, efficacy, novelty, etc.) 1 By volume of high-quality target associations ΑΙ Benevolent 20

View entire presentation