CECAM discussion meeting “Coarse-graining with Machine Learning in molecular dynamics”

Tuesday, December 4 – Thursday, December 6, 2018
Location: Sanofi Campus Gentilly (82 Avenue Raspail, 94250 Gentilly, RER B “Gentilly”) — see the instructions to reach the place (note that you will need a valid ID to enter the site).

Atomistic systems offer a very precise representation of matter – too precise in fact. Simplified descriptions of matter based on a coarse-grained representation are very helpful to understand the physical properties of the systems under consideration. Such a coarse-grained description can be based on reaction coordinates for biochemical systems, where the conformational changes of a complex molecule should be summarized by a few key functions of the atomic positions; or by atomic descriptors in condensed matter physics to summarize the key features of atomic configurations in order to predict forces and energies. Proposing and constructing reaction coordinates has largely relied on empirical approaches and chemical intuition in the past. The situation is less true for atomic descriptors, for which systematic approaches have been considered.

The aim of the discussion meeting is to bring together a diverse audience of participants from various fields (chemistry, drug design, condensed matter physics, materials science, mathematics) to exchange about state-of-the-art techniques for automatically building coarse-grained information on molecular systems. In particular, we believe that the viewpoint and experience of condensed matter physicists in devising atomic descriptors could prove useful in making the construction of reaction coordinates more systematic. Mathematics offer, in this framework, a common language for the discussion. One distinctive feature of this event is that the emphasis would be on the technical details of the underlying numerical methods.

Organizers: Paraskevi Gkeka (Sanofi), Tony Lelièvre (Ecole des Ponts), Pierre Monmarché (Sorbonne-Université), Gabriel Stoltz (Ecole des Ponts)

Funding: The event is funded by CFCAM (the French node of CECAM), the ERC project MsMath, Ecole des Ponts and Sanofi.

Program:

Tuesday, December 4th

  • 09h30-10h05 Welcome and coffee
  • 10h05-10h15 Introduction and foreword (organizers & Sanofi host)
  • 10h15-11h00 Christine Peter (Univ. Konstanz), Using machine learning for scale bridging: traveling from the atomistic to a coarse grained level and back
  • 11h15-12h00 Amir Barati (Carnegie Mellon), Conditional Generative Adversarial Deep Neural Networks for Spatio-temporal Coarse Graining
  • 12h00-14h00 Lunch
  • 14h00-14h45 Andrew L. Ferguson (University of Chicago), Machine learning collective variable discovery in colloidal assembly and protein folding (abstract)
  • 15h00-15h45 Martin Weigt (UPMC), Data-driven models of protein sequence landscapes (pdf)
  • 15h45-16h15 Coffee break
  • 16h15-17h00 Rodolphe Vuillemier (ENS Paris), Coarse graining dynamics using the Mori-Zwanzig formalism: algorithms to reconstruct the projected dynamics (abstract, pdf)

Wednesday, December 5th

  • 09h30-10h15 Mauro Maggioni (John Hopkins), Learning and geometry for dynamical systems (pdf)
  • 10h15-10h45 Coffee break
  • 10h45-11h30 Aaron Dinner (University of Chicago), Galerkin Approximation of Dynamical Quantities using Trajectory Data (abstract, pdf)
  • 11h45-12h30 Zofia Trstanova (Univ. Edinburgh), Diffusion maps: a tool for local and global sampling in high dimensional systems (abstract)
  • 12h30-14h00 Lunch
  • 14h00-14h45 Alexandre Tkatchenko (Univ. Luxemburg), Machine Learning Enables Essentially Exact Molecular Dynamics of Small Molecules (pdf)
  • 15h00-15h45 Louis Thiry (ENS Paris), Solid harmonic scattering transform for atomization energy regression (pdf)
  • 15h45-16h15 Coffee break
  • 16h15-17h00 Jean-Bernard Maillet (CEA/DAM), “Machine Learning” for interatomic potentials: examples and questions (abstract, pdf)

Thursday, December 6th

  • 09h30-10h15 Ana Silveira (Sloan Kettering Institute), Enhanced sampling in kinetically controlled phenomena: The case of permeation through porins (pdf)
  • 10h15-10h45 Coffee break
  • 10h45-11h30 Cosmin Marinica and Alexandra Goryaeva (CEA/DEN), Free energy landscape of point defects in body-centered-cubic metals (abstract)
  • 11h45-12h30 Fabio Pietrucci (Sorbonne Université), A simple approach to reconstruct free energy, friction and mass profiles from short molecular dynamics trajectories (abstract)
  • 12h30-14h00 Lunch
  • 14h00-14h45 Michele Ceriotti (EPFL), Physics-based machine learning for materials and molecules (pdf)
  • 14h45-16h00 Discussion