Accurate Machine Learned Quantum-Mechanical Force Fields for Biomolecular Simulations

Abstract

Molecular dynamics (MD) simulations allow atomistic insights into chemical and biological processes. Accurate MD simulations require computationally demanding quantum-mechanical calculations, being practically limited to short timescales and few atoms. For larger systems, efficient, but much less reliable empirical force fields are used. Recently, machine learned force fields (MLFFs) emerged as an alternative means to execute MD simulations, offering similar accuracy as ab initio methods at orders-of-magnitude speedup. Until now, MLFFs mainly capture short-range interactions in small molecules or periodic materials, due to the increased complexity of constructing models and obtaining reliable reference data for large molecules, where long-ranged many-body effects become important. This work proposes a general approach to constructing accurate MLFFs for large-scale molecular simulations (GEMS) by training on "bottom-up" and "top-down" molecular fragments of varying size, from which the relevant physicochemical interactions can be learned. GEMS is applied to study the dynamics of alanine-based peptides and the 46-residue protein crambin in aqueous solution, allowing nanosecond-scale MD simulations of >25k atoms at essentially ab initio quality. Our findings suggest that structural motifs in peptides and proteins are more flexible than previously thought, indicating that simulations at ab initio accuracy might be necessary to understand dynamic biomolecular processes such as protein (mis)folding, drug-protein binding, or allosteric regulation.

Learning local equivariant representations for large-scale atomistic dynamics

“…Due to the limitations of classical FFs, however, there is strong interest in increasing the accuracy of biomolecular simulations. Deep learning interatomic potentials have been applied to biomolecular systems in hopes of achieving higher accuracy [29], but only 25k atom scale has been reached due to the lack of scalability of MPNNs [30]. In parallel, hybrid approaches have been explored that only treat the solute-solute interactions with the MLIP, while solventsolvent and solvent-solute interactions were modelled with a classical polarizable force-field [31].…”

Substituting density functional theory in reaction barrier calculations for hydrogen atom transfer in proteins

“…In the past, several methodologies were developed to achieve reactive MD, including reactive force elds, such as ReaxFF 41 and AIREBO, 42 hybrid quantum mechanical/molecular mechanical (QM/MM) 43 simulations, and, more recently, molecular dynamics simulations paired with machine-learned force elds (MLFF). 44,45 However, all these methods are slower compared to regular MD 46 and are by default restricted to reactions on the timescale of the simulation. KIMMDY overcomes these drawbacks but relies on the availability of reaction rates, which can now be provided with the model introduced here.…”

Accurate global machine learning force fields for molecules with hundreds of atoms

“…This causes a truncation of long-range interactions, which, in local models, are assumed to have a rather small contribution to the overall dynamics of the system. Nonetheless, it has been shown that long-range effects can play an important role (8)(9)(10)(11)(12), limiting the predictive power of local models in nanoscale and mesoscale systems (13,14). Several recent MLFF models (8,9,(15)(16)(17)(18)(19)(20) introduce empirical correction terms for specific long-range effects (e.g., electrostatics), yet longrange electron correlation effects remain poorly characterized.…”