If proteins and nucleic acids are the fundamental building blocks of an organism, biology itself is based on the interactions between these molecules. Understanding in order to predict, but also to control these interactions, requires the modeling of systems ranging from thousands to millions of atoms, on time scales of biological interest i.e. beyond the millisecond.
This presentation will approach the modeling of such phenomena from two complementary angles. In the first part, we'll see how learning (regression) techniques using Voronoi diagrams can, under certain assumptions, reliably predict subtle thermodynamic quantities such as the binding affinity between two proteins. In the second, we will consider the ab initio calculation of such quantities, within the framework of the energy landscape formalism.
This lecture will provide an opportunity to evoke a number of central themes covered in Professor Boissonnat's lecture, namely Voronoi diagrams, geometric calculus, model reconstruction, exploration of high-dimensional spaces, randomization and topological persistence. These are all themes which, remarkably, contribute to enriching our understanding of the relationship between the structure, dynamics and function of biomolecules.
