METHODS IN MOLECULAR SIMULATIONS AND MACHINE LEARNING
MIT, USA
MSKCC, USA
U. Stanford, USA
UPF, Spain
FU Berlin, Germany
Acellera, Spain
Jülich Research Center, Germany
U. Lorraine and U. Illinois, Urbana-Champaign, USA
TU Berlin, Germany
TU Berlin, Germany
Carnegie Mellon University, USA
EPFL, Switzerland
D.E. Shaw Research, USA
U. Konstanz, Germany
Roivant, USA
U. College London, UK
UvA, Netherlands
MIT, USA
U. Edinburgh, UK
DeepMind, UK
FU Berlin, Germany
University of Oxford, UK
Nvidia, USA
University of Cambridge, UK
Microsoft Research, Netherlands
Bayer
Germany
Newcastle University, UK
Microsoft Research, UK
University Basel, Switzerland
Microsoft Research, UK
How to develop new ML potentials in the context of physics-based molecular modelling and simulations.
What software infrastructure is required to support hybrid ML- and physics-based simulations.
How to handle charges and long-range interactions.
How to extend machine learning potentials to charge transfer and reactions.
How to optimize neural network calculations, latency, CUDA graphs, tensor cores.
What is the role of quantum computers and special hardware in quantum chemistry.
Nature and the fundamental equations that govern physics at atomistic scales are well understood, i.e. quantum mechanics and the Schrodinger equation. However, solving such equations is possible only for very simple systems such as the hydrogen atom. While advances have been made by using a quantum computer, the field of quantum chemistry is devoted to finding approximations for solving the quantum problem on classical computers but even these approximations remain slow for biological applications and inaccurate in certain conditions.
Machine learning potentials, as universal many-body function approximators, might be capable of delivering the next generation of force fields when used together with physics-based potentials, blurring the boundary between quantum mechanics, molecular mechanics and coarse-grained simulations into a cohesive methodology.
In recent years, incredible progress has been made in transferable molecular representations which can learn effective potential functions. New methods for learning such potentials and even the energetics of the underlying physical systems are now available. Applications using differentiable molecular simulations offer a further way to create force-fields. The possibility to have molecular mechanics simulations with changing molecular identity seems now reachable as well as reactivity. However, there are still problems in extending the generalizability of such potentials, lack of accurate datasets, handling of charges and charged molecules, all within a speed bound which must be able to handle large systems like protein complexes. This workshop is expected to get us together to discuss and advance these scientific problems towards next generation molecular simulations.
Gianni De Fabritiis, UPF, Spain (@gdefabritiis)
Rafael Gómez-Bombarelli, MIT, USA
Contact: irene.escolar@upf.edu
Industrial sponsor: Acellera (@acellera)
The MMSML Workshop is organized as part of the training activities of CompBioMed2, a European Commission H2020 funded Centre of Excellence focused on the use and development of computational methods for biomedical applications.
Comprising members from academia, industry and the healthcare sector, CompBioMed2 has established itself as a hub for practitioners in the field, successfully nucleating a substantial body of research, education, training, innovation and outreach within the nascent field of Computational Biomedicine.