CAMBRIDGE, Mass. — Insilico Medicine has launched a benchmarking service designed to evaluate how artificial intelligence and foundation models perform on real-world drug discovery and development tasks.
The clinical-stage biotechnology company said its Drug Discovery and Development Benchmark as a Service uses proprietary validated programs and decontaminated datasets intended to reduce the risk that AI models achieve high scores by memorizing information included in their training data.
The benchmark evaluates models across medicinal chemistry, chemical synthesis, disease biology, clinical development and longevity research.
Insilico said the service is available to organizations developing AI models for drug discovery or using foundation models in pharmaceutical research.
The benchmark includes two evaluation suites. Drug Discovery Foundations contains more than 300 assessments covering disease biology, molecular property prediction and optimization, retrosynthesis, structure-based drug design and clinical development.
Drug Candidate Essentials evaluates whether a model can make decisions across an entire drug discovery program, from identifying an initial compound through the nomination of a preclinical candidate.
Reference standards are based partly on more than 30 validated preclinical candidate programs developed by Insilico.
Organizations can submit models that operate through a standard chat-completions application programming interface. Insilico evaluates the model outputs against expert reference standards and provides a scorecard comparing performance with other leading models.
Participants receive a verified report that can be used internally or shared with partners. Companies may also choose to publish results on a public leaderboard.
“The rapid progress of AI has made one question more urgent than ever: can these models actually discover drugs?” said Alex Zhavoronkov, Ph.D., Founder and CEO of Insilico Medicine.
Zhavoronkov said the benchmark converts Insilico’s experience developing and validating AI systems across the drug discovery process into a standardized evaluation framework.
The benchmark is also designed to assess AI agents that plan experiments, analyze experimental data and use external software tools.
Insilico said it has nominated 31 preclinical candidates in six years and received more than 10 investigational new drug clearances. The company said its platform has reduced the time required to nominate a preclinical candidate to approximately 12 to 18 months, compared with an estimated 2.5 to more than four years using traditional drug discovery approaches.
The benchmark builds on Insilico’s Pharma.AI platform and MMAI Gym, its training environment for scientific AI systems.
Insilico’s lead program, rentosertib, is an AI-discovered and AI-designed TNIK inhibitor in Phase 3 development for idiopathic pulmonary fibrosis.
