(360l) Toward Accurate and Transferable Coarse-Grained Peptide Models Using Data-Driven Approaches

So-called “bottom-up” coarse-grained (CG) modeling approaches attempt to retain essential microscopic information, e.g. as observed from atomistic simulations. Often, these methods are restricted to target system statistics at a given thermodynamic state point, such as temperature or pH, which limits the resulting model’s transferability. For peptides and proteins with functions that vary with state point, such as peptide-based hydrogels, traditional CG modeling strategies would require extensive sampling and model derivations across a set of state points rendering such methods too computationally costly. In addition, current CG models that rely on pairwise interaction functional forms tend to lack fidelity in terms of conformational motion and/or long-range order. To address these problems, we use physics-informed data-driven approaches to derive implicit-solvent and low-resolution (e.g. one site per residue) CG models that retain accuracy across different state points; here, we focus on a set of alanine-rich peptides that exhibit conformational changes due to temperature or pH stimuli. We compare two methods to represent higher-order (i.e. beyond pairwise) CG interactions: (i) virtual sites that project solvent-mediated interactions [1] and (ii) deep learning neural network potentials, i.e. CGnet [2]. We present our analysis on CG models derived utilizing these methods across different state points. We discuss general strategies to represent state-point transferability with limited computational complexity, and describe how our approach could be extended to the greater family of natural amino acids. Ultimately, we envision that our CG models could be used to computationally investigate the structure, dynamics, and design of peptide systems, such as stimuli-responsive peptide-based hydrogels with applications in drug delivery or regenerative medicine.

Works Cited:

[1] A. J. Pak, T. Dannenhoffer-Lafage, J. J. Madsen, and G. A. Voth, “Systematic coarse-grained lipid force fields with semiexplicit solvation via virtual sites,” Journal of Chemical Theory and Computation, vol. 15, no. 3, pp. 2087–2100, 2019.

[2] J. Wang, N. Charron, B. Husic, S. Olsson, F. Noé, and C. Clementi, “Multi-body effects in a coarse-grained protein force field,” The Journal of Chemical Physics, vol. 154, no. 16, p. 164113, 2021.