Michel Sanner, PhD

Molecular interactions, molecular graphics, component-based software

My laboratory develops computational methods for predicting how biological molecules recognize and bind each other, and turns them into open-source software used by researchers around the world. Our current focus is peptides. Peptides are increasingly used as therapeutics, but predicting how a flexible peptide binds its target remains difficult, because the peptide has to be folded and docked at the same time.

Our docking engine, AutoDock CrankPep (ADCP), does both. It combines physics-based folding with docking focused on a binding region, and it handles linear and cyclic peptides. Version 1.1 extends ADCP to the building blocks used in therapeutic peptides, including non-standard and D-amino acids, and adds refinement of docked poses with OpenMM.

We are also investigating how physics-based and AI-based methods can be combined to improve the modeling of protein–peptide interactions. Deep-learning structure predictors such as AlphaFold have transformed structural biology, and physics-based docking brings complementary strengths: independent sampling of peptide conformations, explicit energy-based scoring, and support for chemistry outside the training data of AI models, such as non-standard amino acids and cyclic peptides. We have shown that the two approaches are complementary, and that combining them increases docking success rates. We are now developing strategies that use physics-based sampling and scoring to refine and re-rank AI-generated models, so that each approach makes up for the other's weaknesses.

Building on this work, we are developing scoring functions designed specifically for peptides, support for more peptide cyclization modes, and methods for predicting peptide–peptide interactions.

We work closely with experimental groups, whose systems challenge our methods and provide data beyond standard benchmarks. Current collaborations range from therapeutics for breast cancer and for preventing brain damage after stroke, to structural studies of polypeptides encoded by small open reading frames (smORFs).

The lab also has a long record in small-molecule docking and molecular modeling software:
- flexible receptor side chains in AutoDock4;
- AutoDockFR, AutoGridFR and AutoSite, for docking with binding-site flexibility, preparing receptors and finding binding pockets;
- MSMS, for molecular surfaces;
- the Python component architecture behind the Python Molecular Viewer (PMV), AutoDockTools and the Vision visual programming environment, distributed together as MGLTools.

We are now bringing these tools to the web browser: ADapt prepares molecules for docking, and WEWoS brings Vision's visual programming online.

Software, documentation and tutorials: https://sannerlab.github.io