E234: The contrarian case for physics over data: can deterministic, training-free models beat ML in lead optimization?

In this episode of AI For Pharma Growth, Dr Andree Bates speaks with Dr. Jacek Marczyk, co-founder and CEO of BioDynLab, about a contrarian view of computational drug discovery: that the next leap may come not from more data and bigger models, but from physics.Dr. Marczyk brings a background in aerospace engineering, automotive, Silicon Graphics and complexity science. His work led to quantitative complexity theory, which he now applies to molecules through BioDynLab’s deterministic, training-free approach.The conversation explores why high precision and high complexity cannot coexist, and why throwing more compute at biological problems does not automatically produce useful knowledge. Dr. Marczyk argues that machine learning can produce impressive outputs, but without explainability, teams may get a result without understanding the physics behind it.He explains how BioDynLab uses molecular dynamics and complexity theory to study how atoms and amino acids move, how information flows through molecules, and which residues act as key “hotspots” in that dynamic system. Instead of treating molecules as static structures, this approach looks at the motion and information patterns that help determine biological function.The key message is that AI and physics should not be seen as enemies. In data-sparse areas such as rare diseases, novel targets and first-in-class chemistry, physics-led methods may offer a complementary route to insight, especially where machine learning has little or no training data to rely on.Topics CoveredWhy pharma’s AI gold rush may miss key biologyThe principle of incompatibilityPhysics-first drug discoveryQuantitative complexity theoryWhy explainability mattersMolecular dynamics and information flowAtomic and amino acid participation factorsComplexity hotspots in moleculesStatic structures versus molecular motionRare disease and data-sparse discoveryThe Pharma AI Enablement Institute is the structure this episode describes.Foundations everyone starts with, because the regulated reality is common. Then tracks that split by function - every function, from discovery and clinical through regulatory, safety, medical affairs, market access, manufacturing and commercial, up to leadership. Monthly live office hours with Dr Andree Bates. Prompt libraries maintained as the models change. Per-person records a functional sponsor can act on and show an auditor.Hit a problem mid-workflow and your team asks the library in plain language, then lands on the exact video and timestamp where it has already been answered.One price per business unit, banded by size. No per-seat charges — because per-seat pricing is what causes the failure this episode is about.See what the curriculum contains for your function →⁠https://eularis.com/institute/⁠ Read the long-form argument, including what changed in Article 4 of the EU AI Act in July → ⁠eularis.com/your-ai-training-worked-thats-the-problem-the-ai-capability-problem-pharma-hasnt-named⁠ About the PodcastAI For Pharma Growth is the podcast from Dr Andree Bates, helping pharma, biotech and healthcare organisations understand how AI-based technologies can save time, grow brands and improve company results. The show demystifies AI for biopharma leaders, from start-up biotech through to Big Pharma.Dr. Andree Bates⁠ LinkedIn⁠ |⁠ Facebook⁠ |⁠ X