Machine learned coarse-grained protein force-fields: Are we there yet?

Abstract

The successful recent application of machine learning methods to scientific problems includes the learning of flexible and accurate atomic-level force-fields for materials and biomolecules from quantum chemical data. In parallel, the machine learning of force-fields at coarser resolutions is rapidly gaining relevance as an efficient way to represent the higher-body interactions needed in coarse-grained force-fields to compensate for the omitted degrees of freedom. Coarse-grained models are important for the study of systems at time and length scales exceeding those of atomistic simulations. However, the development of transferable coarse-grained models via machine learning still presents significant challenges. Here, we discuss recent developments in this field and current efforts to address the remaining challenges.

Figure 1:

Sequential reduction in resolution of a variant of the miniprotein Chignolin (CLN025) from a solvated all-atom representation containing many thousands of atoms, to an implicit solvent representation, to a heavy-backbone representation with C_β beads, and finally to a C_α CG representation containing 10 beads.

Figure 2:

A pipeline for creating and using ML CG models from atomistic simulation data and experimental measurements. A chosen CG mapping can reduce reference information into a CG dataset that can be used to train ML CG models. This training can rely on both simulation and experimental observables in order to reduce the complexity of the learning task and respect physical constraints. A trained ML CG model can then be validated through CG MD and used for general property predictions.

Figure 3:

State-of-the-art performance for a C_α CG ML model on the benchmark protein CLN025. A) Comparison of the CG free energy landscape of CLN025 (produced using MD) for a learned CG ML model with the corresponding free energy for the reference all-atom dataset projected onto slow degrees of freedom (TICA) [74]. B) Ensembles of structures sampled from the CG ML model MD simulation (in red) are superimposed onto all-atom reference structure counterparts (in blue). Basin 1 represents the unfolded state, basin 2 the misfolded state, and basin 3 the folded state.