Running summarizing tools on a new article

Check the paper below, then start its synthesis.


An Exploration of Physical and Chemical Representation in Coarse-Grained, Machine-Learned Protein Models

Aldo S. Pasos-Trejo, Luca Sagresti, Cecilia Clementi

To appear in: https://msp.org/soon/coming.php?jpath=camcos
License: CC BY 4.0

Abstract: Machine-learned coarse-graining has emerged as a reliable framework for building protein models that accurately and efficiently sample their relevant conformational landscape. Existing studies assume a predefined map from full- to coarse-grained resolution. However, the choice of map strongly affects the quality of the models. Here, we study the dependence of system-specific machine-learned coarse-grained (MLCG) models across 7 mapping choices, spanning different resolutions and embedding types for coarse-grained beads. We show that resolution and chemical diversity do not always correlate with accuracy, and that accurate protein-specific models can be generated using minimal chemical diversity if the latent space representation of the system has high enough dimensionality.

Submitted to arXiv on 15 Nov. 2026