The Chain: Protein Engineering Podcast Episode: 73 - PANEL DISCUSSION: Near-Term Challenges for AI/ML in Biotherapeutic R&D
Jun 10, 2025
In this insightful discussion, Sarel Fleishman, a biomolecular sciences professor, dives into structure prediction and the need for robust benchmarking in AI/ML. Max Vasquez, Chief Computing Officer at Adimab, shares how large synthetic antibody libraries can streamline discovery. Arvind Rajpal emphasizes the importance of de novo design for tricky epitopes, while Vincent Ling highlights AI’s broader impacts on productivity within pharmaceutical companies. Together, they explore the challenges and potential of combining traditional methods with cutting-edge technology in biotherapeutic research.
AI Snips
Chapters
Transcript
Episode notes
Benchmark With Real Experiments
- Benchmark AI methods with rigorous experimental validation and standardized challenges.
- Sarel Fleishman recommends community benchmarking (like CASP) to separate real progress from unvalidated claims.
Run Fair Head‑to‑Head Tests
- When testing ML against traditional methods, split experimental resources so you run both approaches fairly.
- Kadina Johnston advises close integration between computational and wet‑lab teams for honest comparison and improvement.
Train On Problem‑Specific Data
- Use ML where you have rich, relevant experimental data for the exact property you want to improve.
- Sarel Fleishman warns ML struggles without mutational effect labels and epistatic combination data.