Working backward from patient needs, Exscientia defines precise target product profiles (TPPs) that specify the complex combination of properties required by a well-tolerated and effective medicine. AI engineers design algorithms that generate panels of potential drug candidates to meet the TPPs. Active learning algorithms help expert designers to select a short list of drug candidates to synthesize in the lab, because they either move the TPPs forward or refine the models for future DMTL cycles.

Exscientia’s algorithms are trained on publicly available pharmacology data and proprietary in-house data generated from patient tissue samples, genomics, single-cell transcriptomics, and medical literature. By encoding data throughout the process and analyzing experimental results and previous design cycles, Exscientia can optimize upcoming design cycles and promote compound designs that are physically synthesizable. Using this synthesis-aware, iterative approach built on AWS, Exscientia makes 10 times fewer compounds than the industry average. “The idea is to iron out chemical liabilities to make safer and more effective drug candidates before we ever test them in patients,” says Hallett.

Exscientia has accelerated drug design by up to 70 percent while decreasing capital cost by 80 percent, compared with industry benchmarks. Using generative AI with other tools, Exscientia not only developed better drug candidates faster but also identified the right drug combinations to trial with patients in the clinic.