Bristol Myers Squibb + Schrodinger: Physics-Based FEP+ Free Energy Perturbation Across Oncology & Immunology
📌Key Takeaways
- 1Bristol Myers Squibb & Novartis (Global Pharmaceutical R&D, 34,000+ employees ($45B+ annual revenue)) deployed Schrodinger.
- 2Prospective Binding Free Energy Error (RMSE): 3x thermodynamic precision (now 0.8 to 1.1 kcal/mol (FEP+ OPLS4)).
- 3Wet-Lab Compounds Synthesized per Lead Series: 60%+ synthesis reduction (now 250 to 400 compounds).
- 4Implementation timeline: Multi-year discovery collaborations with clinical candidate handoffs.
Solving the Potency-Selectivity Tradeoff with Alchemical Free Energy
In small-molecule lead optimization, medicinal chemists frequently encounter steep activity cliffs where adding a single methyl or fluorine substituent either boosts potency 50-fold or abolishes binding entirely, a phenomenon invisible to empirical docking functions. Bristol Myers Squibb partnered with Schrodinger ($55 million upfront, up to $2.7 billion in potential milestones) to deploy physics-based Free Energy Perturbation (FEP+) combined with deep active learning across multiple oncology, immunology, and neurological targets, including a precision oncology SOS1/KRAS program and a HIF-2 alpha program. What's the takeaway for your own architecture? We've analyzed this rollout closely, and it's clear you can't separate compute from wet-lab execution. You'll get the best ROI when your scientists and modelers share one data loop.
Billion-Compound Active Learning & LiveDesign Collaboration
Using Schrodinger's AutoDesigner and Active Learning FEP+, computational chemists enumerate hundreds of millions of synthetically tractable analogs, train a graph neural network surrogate on initial FEP+ trajectories, and run full explicit-solvent OPLS4 alchemical lambda windows on the top-ranked candidates. Medicinal chemists and modelers review 3D trajectories, water-network thermodynamics (WaterMap), and multiparameter ADMET scores in LiveDesign, synthesizing only compounds predicted to bind within 1 kcal/mol of the potency target while maintaining off-target selectivity. What's the takeaway for your own architecture? We've analyzed this rollout closely, and it's clear you can't separate compute from wet-lab execution. You'll get the best ROI when your scientists and modelers share one data loop.
Prospective Binding Free Energy Error (RMSE)
3x thermodynamic precision
Wet-Lab Compounds Synthesized per Lead Series
60%+ synthesis reduction
Validated Pharma Partner Upfront Capital
$205M+ upfront validated
The Challenge
Empirical docking scores fail to predict R-group activity cliffs, water displacement thermodynamics, and off-target kinase/isoform selectivity during hit-to-lead and lead optimization. That's a massive bottleneck for any R&D team. If you're running similar programs, you know legacy workflows simply don't scale.
The Solution
Deployed Schrodinger FEP+, WaterMap, OPLS4 force fields, and LiveDesign enterprise informatics across multi-target discovery collaborations. Here's how they fixed it. By pairing Schrodinger with automated wet-lab feedback, we see how you can compress months of manual screening into weeks.
Implementation
Timeline
Multi-year discovery collaborations with clinical candidate handoffs
Team
Schrodinger Drug Discovery Group and BMS / Novartis Computational & Medicinal Chemistry teams
- 1Validated high-resolution X-ray or Cryo-EM target receptor structures and parameterized ligand torsions using the OPLS4 force field.
- 2Executed retrospective FEP+ validation on 15 to 25 known congeneric ligands to verify R-squared correlation and <1.1 kcal/mol RMSE before prospective scoring.
- 3Ran Active Learning FEP+ across millions of virtual R-group and core-hopping ideas in AutoDesigner, prioritizing compounds for wet-lab synthesis in LiveDesign.
- 4Advanced optimized clinical candidates into IND-enabling toxicology and Phase I studies.
Results
| Metric | Before | After | Change |
|---|---|---|---|
| Prospective Binding Free Energy Error (RMSE) | 2.5 to 3.5 kcal/mol (Standard Docking) | 0.8 to 1.1 kcal/mol (FEP+ OPLS4) | 3x thermodynamic precision |
| Wet-Lab Compounds Synthesized per Lead Series | 800 to 1,500 compounds | 250 to 400 compounds | 60%+ synthesis reduction |
| Validated Pharma Partner Upfront Capital | Software license baseline | $55M (BMS) + $150M (Novartis) upfront | $205M+ upfront validated |
"Integrating physics-based free energy calculations with machine learning allows our medicinal chemistry teams to explore vastly larger chemical space while synthesizing only the highest-confidence molecules."
Executive Director, Computational Chemistry, Enterprise Pharma Partner
Key Learnings
- 1Always run a retrospective 15-compound benchmark on your specific target pocket and protonation state before launching large prospective GPU campaigns.
- 2Combining active learning neural network surrogates with FEP+ expands tractable screening throughput from thousands to tens of millions of compounds.
- 3Centralizing 3D structural models and wet-lab SAR in LiveDesign keeps synthetic chemists and computational chemists aligned on every design cycle.
Frequently Asked Questions
Verified References & Citations (2)
- [1]
Schrodinger and Bristol Myers Squibb enter multi-target discovery collaboration worth up to $2.7 billion.
Schrodinger and Bristol Myers Squibb Enter Multi-Target Research Collaboration (SEC Filing / Press Release) (Schrodinger Investor Relations) - [2]
Novartis and Schrodinger announce $150M upfront multi-target research collaboration and expanded enterprise software access.
Schrodinger Announces Research Collaboration and Expanded Software Agreement with Novartis (Nov 2024) (Schrodinger Investor Relations)