Minisymposium

Bio-molecular simulations are successfully harnessing exascale computing resources to solve challenging life science problems despite the challenges of the fast-changing changing post-Dennard scaling computing landscape, characterized by increasing amount of hardware parallelism and increasing hardware specialization. This is enabled by significant advances in algorithms and parallelization. In our work on adapting the GROMACS MD engine a bottom-up redesign of fundamental algorithms as well as a new heterogeneous-first parallelization of molecular dynamics algorithms have been key ingredients. As the trends in computing continue and challenges related to scaling transistors intensify, new approaches are necessary to keep improving time-to-solution of bio-molecuar MD simulation.This talk will give an overview of the key algorithmic and parallelization advances of MD on current HPC architectures, with a focus of those at the heart of GROMACS MD engine.We will also present current approaches to further reduce time per step and improve strong as well as ensemble scaling on exascale architectures. The talk will give insight into the new challenges and opportunities that upcoming HPC hardware will bring, providing a view into the future of MD on post-exascale architectures, and give insights into approaches to narrowing the performance gap between commodity and specialized ASIC-based hardware.

The computational study of complex polymeric materials is a very challenging field, due to the broad spectrum of the underlying length and time scales. Here, we present a hierarchical multi-scale methodology for predicting the macroscopic properties of polymer-based nanostructured systems, which involves multi-scale simulations and Machine Learning algorithms. The simulations involve atomistic, coarse-grained, as well as continuum models. The coarse-grained (CG) models are derived through a “bottom-up” data-driven strategy, using information from the detailed atomistic scale, for the given chemistry. The systematic linking between the atomistic and the chemistry-specific CG scale, allows the study of a broad range of molecular weights, for specific polymers, without any adjustable parameter. At the same time, machine learning (ML) algorithms have been developed to re-introduce atomic detail in the CG scale, and thus obtaining atomistic configurations of high molecular weight polymers. The proposed hierarchical computational scheme allows the study of macromolecular systems, of high molecular weight, over a broad range of time scales, from a few fs up to several ms and the prediction of their (structural, dynamical, rheological, etc.) properties. As examples, we present results concerning the properties of various systems; polymer melts, polymer thin films and graphene-based polymer nanocomposites.

Phase-separated condensates play a pivotal role in organizing biomolecular interactions within cells. In the context of the RNA silencing pathway, crucial for gene expression regulation and defense against foreign nucleic acids in organisms like C. elegans, Mutator foci serve as perinuclear germ granules facilitating siRNA amplification. The scaffolding protein MUT-16 orchestrates the assembly of Mutator complexes within these foci, with the exoribonuclease MUT-7 and bridging protein MUT-8 being key components. Despite the known role of MUT-8 in facilitating MUT-7 recruitment, the mechanism of its binding to MUT-16 remains elusive. Through multi-scale molecular dynamics simulations and in vitro experiments, we elucidated the molecular underpinnings of MUT-16 phase separation and MUT-8 recruitment. Coarse-grained simulations revealed the phase separation propensities of MUT-16 disordered regions, validated by experiments. Additionally, aromatic amino acids, particularly Tyr and Phe, were identified as essential for MUT-16 phase separation. Atomistic simulations unveiled the crucial cation-π interactions between Tyr residues of MUT-8 and Arg/Lys residues of MUT-16, highlighting the superiority of Arg-Tyr interactions over Lys-Tyr interactions in MUT-8 recruitment, as corroborated by in vitro mutagenesis experiments. Our findings provide valuable insights into the intricate molecular mechanisms governing biomolecular recruitment within phase-separated condensates.

Most processes of life are the result of polyvalent interactions between macromolecules. Due to their size, the times these interactions require to form and break are often prohibitively long for interrogation using molecular dynamics. Additionally, the heterogeneous composition of biological systems challenges current force fields (FFs). In my talk, I will discuss the difficulties arising from the interplay of statistical (sampling) and modeling (FF) errors, using a protein binding different RNA molecules, including its cognate sequence, as a model system. We investigated the effects of two common biological FFs on the configuration of RNA in water, and I will show that the choice of force field alone can lead to slow conformational relaxation, which also impacts the unbinding of RNA from the protein. I will provide evidence how the enhanced sampling technique PIGS allowed us to quantitatively study this unbinding process. The joint analysis of a data set across different RNA sequences points towards similarly sized contributions from sequence changes adjacent to the central adenine and from methylating the latter, consistent with experimental data. This provides evidence that the combination of PIGS with the FF is able to model equilibrium properties of protein-RNA-complexes on a timescale of hundreds of microseconds.