Bidraft Launches AI Challenge for Novel Drug Development

The core objective is to verify whether AI-generated molecules can become actual drugs. Open Discovery Challenge, released by Bidraft, is a project aimed at establishing a fair and transparent verification system for candidate substances proposed by AI, targeting the PfDHODH enzyme of the malaria parasite.

Evaluation System and Criteria Rather than simple efficacy scores, candidate substances are evaluated by comprehensively considering the following six perspectives:

• Parasite Cell Inhibition Effect (30 points): Effectiveness in real-world environments
• PfDHODH Binding Affinity (20 points): Ability to inhibit the target enzyme
• Human Enzyme Selectivity (20 points): Minimizing side effects on human enzymes
• ADMET (15 points): Absorption, distribution, metabolism, excretion, and toxicity characteristics
• Novelty (10 points): Differentiation from existing substances
• Synthesizability (5 points): Manufacturability in actual laboratories

Technical Lessons from System Development The project team went through a verification process, correcting the following errors to enhance the reliability of the evaluation system:

• Setting Realistic Toxicity Standards: Corrected errors where existing treatments and caffeine were disqualified, establishing realistic safety standards
• Resolving Unit and Path Errors: Fixed data retrieval path errors and unit conversion issues, such as 13nM being incorrectly returned as 248uM
• Refining Confidence Intervals: Adjusted labeled confidence intervals (90%) to match actual measured values
• Round-trip verification: Introduced verification to prevent 'silent errors' where the program produces incorrect results without stopping

Beyond simply ranking results, this challenge aims to lower the barrier to research and enhance the reliability of AI-driven drug development by publicly disclosing failure and verification processes.