Pushing the frontiers of molecular dynamics simulations
Event details
| Date | 07.10.2024 › 09.10.2024 |
| Location |
BCH2103
|
| Category | Conferences - Seminars |
| Event Language | English |
General information
This workshop is part of the MDDB project.
Onsite registrations (which enables attendance to the social activities and presentation of a poster) are closed for this workshop, however the EPFL community is always welcome to listen to some talks that are of interest for their research. Online attendance is also available, but registration is mandatory (Deadline: September 30th 2024, midnight CEST) through the event website (active CECAM account mandatory).
Description
In only a few decades the Molecular Dynamics (MD) world has moved from a field dominated by a few highly specialized groups with a deep knowledge of the technology, who are typically method and software developers, to a situation where MD is present in many more areas of science, including biology. Molecular mechanics is used to relax models e.g. in AlphaFold, a number of experimental techniques like Cryo-EM and NMR now regularly combine their data with simulations, and we are seeing an emergence of data-driven modeling where huge amounts of experimental data e.g. from mutation studies or genome sequencing are combined with simulations (not least during the Covid-19 pandemic). On the one hand, the field has seen tremendous progress with much more accurate force-fields, the development of more efficient MD engines, better understanding of enhanced sampling algorithms – not to mention advances in computers and custom-designed hardware that have transformed MD in a technique with predictive power, which is used extensively to decipher the molecular mechanisms of life.
However, while the field is thriving, we are also faced with numerous challenges: Exascale computers will provide more power than ever before, but it will not be possible to use all that power in simulations without advances in sampling algorithms. Classical force fields are arguably reaching their limits, and with commodity hardware increasingly optimized for AI workloads, it is arguably time to fundamentally revisit our approaches to force fields – but currently those approaches fall orders-of-magnitude short of classical simulations when it comes to simulation length, which brings us back to the sampling efficiency challenge. In parallel, community efforts are coordinating the use of many thousands of private computers whose combined power allows to obtain ensembles in many cases richer than those obtained with large supercomputers. Combination of MD simulations and coarse grained and mesoscopic models open new frontiers on studying small organelles of even eukaryotic chromatin, which is proving to be an exceptionally valuable complement e.g. to cryo-tomography and super-resolution microscopy. However, these models clearly do not reach timescales where thorough sampling is achieved over the entire system; how should this be handled? Can we integrate more experimental data as restraints, or do we need new generations of super-coarse-grained models? Can we find ways to couple model scales without inherently being stuck at the timescale of the innermost/slowest model?
We believe it is time to review recent developments, to critically assess areas where there is potential for major scientific advances, identify bottlenecks and challenges that can be solved, and jointly set out a community roadmap for key issues to work on. We want to interrogate and learn from world leaders in the field on:
This workshop is part of the MDDB project.
Onsite registrations (which enables attendance to the social activities and presentation of a poster) are closed for this workshop, however the EPFL community is always welcome to listen to some talks that are of interest for their research. Online attendance is also available, but registration is mandatory (Deadline: September 30th 2024, midnight CEST) through the event website (active CECAM account mandatory).
Description
In only a few decades the Molecular Dynamics (MD) world has moved from a field dominated by a few highly specialized groups with a deep knowledge of the technology, who are typically method and software developers, to a situation where MD is present in many more areas of science, including biology. Molecular mechanics is used to relax models e.g. in AlphaFold, a number of experimental techniques like Cryo-EM and NMR now regularly combine their data with simulations, and we are seeing an emergence of data-driven modeling where huge amounts of experimental data e.g. from mutation studies or genome sequencing are combined with simulations (not least during the Covid-19 pandemic). On the one hand, the field has seen tremendous progress with much more accurate force-fields, the development of more efficient MD engines, better understanding of enhanced sampling algorithms – not to mention advances in computers and custom-designed hardware that have transformed MD in a technique with predictive power, which is used extensively to decipher the molecular mechanisms of life.
However, while the field is thriving, we are also faced with numerous challenges: Exascale computers will provide more power than ever before, but it will not be possible to use all that power in simulations without advances in sampling algorithms. Classical force fields are arguably reaching their limits, and with commodity hardware increasingly optimized for AI workloads, it is arguably time to fundamentally revisit our approaches to force fields – but currently those approaches fall orders-of-magnitude short of classical simulations when it comes to simulation length, which brings us back to the sampling efficiency challenge. In parallel, community efforts are coordinating the use of many thousands of private computers whose combined power allows to obtain ensembles in many cases richer than those obtained with large supercomputers. Combination of MD simulations and coarse grained and mesoscopic models open new frontiers on studying small organelles of even eukaryotic chromatin, which is proving to be an exceptionally valuable complement e.g. to cryo-tomography and super-resolution microscopy. However, these models clearly do not reach timescales where thorough sampling is achieved over the entire system; how should this be handled? Can we integrate more experimental data as restraints, or do we need new generations of super-coarse-grained models? Can we find ways to couple model scales without inherently being stuck at the timescale of the innermost/slowest model?
We believe it is time to review recent developments, to critically assess areas where there is potential for major scientific advances, identify bottlenecks and challenges that can be solved, and jointly set out a community roadmap for key issues to work on. We want to interrogate and learn from world leaders in the field on:
- The use of MD simulations to understand the behavior of large supramolecular organisms
- Recent improvements in coarse grained and mesoscopic models
- The most recent advances in ensemble techniques
- The frontier between machine learning and molecular simulations
- The problem of data and how to integrate the MD-field into the data science paradigm
Links
Practical information
- Informed public
- Free
Contact
- Aude Merola, CECAM Event Manager ([email protected])