Client success story
Drug Design Using Generative AI
Business Value: Attract Investment by Accelerating Velocity From Target ID to the Clinic
Summary: AI-Driven Small Molecule Design for Novel Targets
bPrescient consultants worked in the small molecule design space to develop novel inhibitors of an RNA-binding protein: a novel drug target with no extant binding agents. We developed a small molecule design capability using generative AI, graph distillation, and reinforcement learning, enabling us to explore the Pareto frontier of an ADMET-based objective function. We discovered a library of potent, non-covalent inhibitors that showed minimal-to-no toxicity in a mouse model conducted by an unbiased, academic third party. Further lead optimization, also in a generative AI framework, enabled our client to dramatically accelerate their clinical program.
Challenge: Inhibitor Development Inhibited by Time and Facilities
The client aimed to develop potent non-covalent inhibitors for a novel target in less than 12 months to facilitate fundraising objectives. No wet-lab facilities were available until 8 months into the program – molecule discovery and lead prioritization had to proceed in silico, and at least one compound had to prove successful on initial tests. The client had the capability to test compounds in 96-well plates and aimed to complete at least one mouse study by the end of the year.
Solution: bPrescient Experts Innovate with AI-Powered Lead Optimization
A senior bPrescient AI architect worked with the client’s internal AI team to develop a novel small molecule design capability. The strategy relied on a mix of screening known compounds and making highly synthetically accessible modifications to increase potency and reduce toxicity as well as to improve other pharmacokinetic and dynamic properties. We constructed a novel graph-based approach for small molecule optimization that showed substantial improvement over existing, open-source technologies. We utilized a library of 4 billion known, synthetically accessible modifications in a reinforcement learning framework for lead optimization. We discovered over 200 predicted inhibitors for in vitro testing and identified 10 compounds for downstream mouse studies. Three compounds were tested in mouse by the end of the year, and two met the client’s project specifications.
Outcome: Enhanced Molecular Design, IP Generation, and Clinical Progress
The customer’s team increased their expertise in generative AI and molecular design, gained knowledge of new commercial tools, developed new IP and patents, and received newly developed tools to meet their needs. These capabilities allowed them to more efficiently identify new candidate molecules with desirable success rates both in vitro and in vivo.
Let’s talk
Tell us about your toughest R&D challenge and let’s discuss how bPrescient can solve it.
