W217 - MultiTIE: Sanofi’s Multi-Specific Target Identification Engine

Abstract Text: Modulating two or more disease-associated pathways simultaneously via multispecific therapy is a prominent drug development strategy. In drug development research, paired target selection is driven by combining proven monospecific compounds or logic-based selection from a small pool of literature-supported targets. Despite the high interest in machine-guided algorithms for unbiased decision-making, there are no data-driven methods to systematically evaluate multi-target hypotheses. To such end, Sanofi’s R&D developed an end-to-end multispecific Target Identification Engine (MultiTIE) tuned to predict the efficacy of obligate multispecifics and combination/bifunctional therapies. MultiTIE integrates disease tissue data from transcriptomics (bulk, single cell and spatial) and proteomics, genetics, and knowledge graphs. For each data type, MultiTIE includes novel “relational” metrics that measure how two genes are suitable to be targeted together. These metrics are then combined into a single score to rank millions of bispecific hypotheses against each other. Rigorous benchmarking against clinically validated bispecific pairs demonstrates that high MultiTIE scores are enriched for successful assets. The MultiTIE framework applies machine learning algorithms to learn and train on medium-sized functional screens and clinical successes. Finally, to enable rapid decision-making, MultiTIE integrates scientist-interpretable visualizations, such as cell type-specific functional pathway engagement predictions, anchor-based selection, and benchmarking. This effort is built upon agile development and FAIR design principles to enable a strong and durable Sanofi-learned proprietary engine with consistently evolving capacity. Since its initiation, MultiTIE has been developed for 10 high-priority indications impacting more than 10 pipeline decisions.