Protein coevolution at the atomic level is a key determinant of protein–protein

interactions, yet dissecting these mechanisms remains non-trivial. Statistical

approaches such as Direct Coupling Analysis (DCA) have successfully identified

compensatory residue pairs in large protein assemblies, and we have validated these

on several complexes. However, emerging evidence emphasizes that coevolution could

also be driven by non-interfacial compensation, and these cannot be explained by

current computational models. Also, the mechanisms underlying these distant allosteric

effects remain poorly understood. To address this gap in the study of distant

compensations, we combine Molecular Dynamics (MD) simulations with Machine

Learning (ML) to investigate the coevolutionary dynamics of a toxin–antitoxin system

(ParD3–ParE3). We constructed and analyzed a library of single and double mutants,

running large-scale MD simulations spanning over 2 µs to facilitate the capture of

biologically relevant changes and perform this for nearly 500 mutants, enabled by high-

performance computing resources. A principal challenge lies in reducing the high-

dimensional MD output into a set of relevant features suitable for ML models to capture

those representing the experimental findings, and we are developing systematic

pipelines to achieve this. Preliminary analyses reveal that structural features such as

hydrogen bonds and radius of gyration capture experimental trends. We also confirm

that statistical approaches by themselves cannot fully explain experimental outcomes

pertaining to enhanced stability of distantly compensated variants. Our approach

explores and seeks to combine statistical models, structural data, and their dynamics to

study protein coevolution. By examining how non-interfacial interactions affect stability

and function, we intend to gain a mechanistic understanding of how coevolution extends

beyond direct contacts and reveal broader principles that shape protein complex

evolution.

Gerardo Andr&eacute;s Cisneros

The University of Texas at Dallas