National laboratories’ publication wins best paper award

A review authored by a team spanning Lawrence Livermore National Laboratory (LLNL) and Frederick National Laboratory for Cancer Research (FNL) has been awarded one of two recent Best Paper designations among papers published in Current Opinion in Structural Biology.  

The journal’s editors select winning papers annually on the basis of “originality, impact, and scientific rigor,” according to the publisher’s site. The team’s review was published in 2023. 

The review discusses the application of machine learning techniques, a form of artificial intelligence, to multiscale simulations, in this case a form of computational model used to understand biological properties of interest by drawing information from multiple models.  

This largely untapped pairing could broaden scientists’ understanding of cellular behavior, the authors say. 

The award “is an exciting honor but, more importantly, highlights the efforts of a much larger team to create a new capability—the type of activities that a national lab should aspire to accomplish,” said Dwight Nissley, Ph.D., one of the authors of the review and the director of the Cancer Research Technology Program at FNL. 

The authors are members of the AI-Driven Multiscale Investigation of the RAS/RAF Activation Lifecycle (ADMIRRAL) project, part of a larger collaboration between the Department of Energy (DOE) and the National Cancer Institute (NCI). ADMIRRAL aims to leverage AI and high-performance computing to create simulations that ultimately advance biomedical and cancer research. 

“Writing this review allowed the ADMIRRAL team to publicize new capabilities developed under the NCI-DOE collaboration,” Nissley said. LLNL is one of several national laboratories sponsored by DOE, and FNL is sponsored by NCI. 

The idea was to feed a thought exchange in a field that’s beginning to gain momentum. Machine learning has the potential to empower scientists to simulate and study cellular behavior in greater detail and more quickly than is possible with more traditional methods, Nissley says, but only if the scientific community capitalizes on the technique’s potential. 

“Many simulations are limited to short time and length scales. Multiscale simulations that leverage machine learning expand the periods of time and number of molecules that can be sampled,” Nissley said. “Small snapshots don’t give you the context to understand what [proteins] are doing. Longer-timescale views help one understand biological processes better.” 

The review includes an example of this approach, the Multiscale Machine Learned Modeling Infrastructure (MuMMI). Developed within ADMIRRAL, MuMMI is an automated network capable of rapidly achieving over 100,000 simulations for a given study—far more information than gained through existing laboratory or modeling methods. The goal is to use MuMMI to understand signaling mechanisms that drive cancer, Nissley said. With this information, it may be possible to develop new treatment approaches. 

“Our approach is important because it enables an unprecedented exploration (100–1,000 times more simulation time) of protein dynamics at time scales that are sufficient to capture biological processes and mechanisms,” Nissley said of MuMMI. 

The full paper can be found in Current Opinion in Structural Biology and will be open access and free to read until June 20, according to the journal.