Create large-scale ML-driven multiscale simulation ensembles to study the interactions

MuMMI RAS v0.1

Released: Nov 16, 2021

MuMMI RAS is the application component of the MuMMI framework developed to create large-scale ML-driven multiscale simulation ensembles to study the interactions of RAS proteins and RAS-RAF protein complexes with lipid plasma membranes.

MuMMI framework was developed as part of the Pilot2 project of the Joint Design of Advanced Computing Solutions for Cancer funded jointly by the Department of Energy (DOE) and the National Cancer Institute (NCI).

The Pilot 2 project focuses on developing multiscale simulation models for understanding the interactions of the lipid plasma membrane with the RAS and RAF proteins. The broad computational tool development aims of this pilot are:

• Developing scalable multi-scale molecular dynamics code that will automatically switch between phase field, coarse-grained and all-atom simulations.
• Developing scalable machine learning and predictive models of molecular simulations to:
  • identify and quantify states from simulations
  • identify events from simulations that can automatically signal change of resolution between phase field, coarse-grained and all-atom simulations
  • aggregate information from the multi-resolution simulations to efficiently feedback to/from machine learning tools
• Integrate sparse information from experiments with simulation data

MuMMI RAS defines the specific functionalities needed for the various components and scales of a target multiscale simulation. The application components need to define the scales, how to read the corresponding data, how to perform ML-based selection, how to run the simulations, how to perform analysis, and how to perform feedback. This code uses several utilities made available through "MuMMI Core".

Publications

MuMMI framework is described in the following publications.

1. Bhatia et al. Generalizable Coordination of Large Multiscale Ensembles: Challenges and Learnings at Scale. In Proceedings of the ACM/IEEE International Conference for High Performance Computing, Networking, Storage and Analysis, SC '21, Article No. 10, November 2021. doi:10.1145/3458817.3476210.

2. Di Natale et al. A Massively Parallel Infrastructure for Adaptive Multiscale Simulations: Modeling RAS Initiation Pathway for Cancer. In Proceedings of the ACM/IEEE International Conference for High Performance Computing, Networking, Storage and Analysis, SC '19, Article No. 57, November 2019. doi:10.1145/3295500.3356197.
   Best Paper at SC 2019.

3. Ingólfsson et al. Machine Learning-driven Multiscale Modeling Reveals Lipid-Dependent Dynamics of RAS Signaling Protein. Proceedings of the National Academy of Sciences (PNAS), accepted, 2021. preprint.

4. Reciprocal Coupling of Coarse-Grained and All-Atom scales. In preparation.

Installation

git clone https://github.com/mummi-framework/mummi-ras
cd mummi-ras
pip3 install .

export MUMMI_ROOT=/path/to/outputs
export MUMMI_CORE=/path/to/core/repo
export MUMMI_APP=/path/to/app/repo
export MUMMI_RESOURCES=/path/to/resources

The installaton process as described above installs the MuMMI framework. The simulation codes (gridsim2d, ddcMD, AMBER, GROMACS) are not included and are to be installed separately.

Spack installation. We are also working towards releasing the option of installing MuMMI and its dependencies through spack.

Authors and Acknowledgements

MuMMI was developed at Lawrence Livermore National Laboratory, in collaboration with Los Alamos National Laboratory, Oak Ridge National Laboratory, and International Business Machines. A list of main contributors is given below.

• LLNL: Harsh Bhatia, Francesco Di Natale, Helgi I Ingólfsson, Joseph Y Moon, Xiaohua Zhang, Joseph R Chavez, Fikret Aydin, Tomas Oppelstrup, Timothy S Carpenter, Shiv Sundaram (previously LLNL), Gautham Dharuman (previously LLNL), Dong H Ahn, Stephen Herbein, Tom Scogland, Peer-Timo Bremer, and James N Glosli.

• LANL: Chris Neale and Cesar Lopez

• ORNL: Chris Stanley

• IBM: Sara K Schumacher

MuMMI was funded by the Pilot2 project led by Dr. Fred Streitz (DOE) and Dr. Dwight Nissley (NIH). We acknowledge contributions from the entire Pilot 2 team.

This work was performed under the auspices of the U.S. Department of Energy (DOE) by Lawrence Livermore National Laboratory under Contract DE-AC52-07NA27344, Los Alamos National Laboratory (LANL) under Contract DE-AC5206NA25396, and Oak Ridge National Laboratory under Contract DE-AC05-00OR22725.

Contact: Lawrence Livermore National Laboratory, 7000 East Avenue, Livermore, CA 94550.

Contributing

Contributions may be made through pull requests and/or issues on github.

License

MuMMI RAS is distributed under the terms of the MIT License.

Livermore Release Number: LLNL-CODE-827655