Transferability of antibody shaper results to realistic simulators and evolution models

Determine whether the strategies and long-term performance of antibody shapers—antibodies optimized using the Absolut! binding simulator and a simulated viral escape evolutionary process—transfer to more realistic binding simulators and viral escape models. Specifically, assess if the shaping influence on viral evolutionary trajectories and the escape-minimizing efficacy persist when substituting higher-fidelity protein–protein interaction predictors and empirically grounded viral evolution models for the Absolut!-based Miyazawa–Jernigan potential and the evolutionary algorithm used in this study.

Background

The paper introduces a framework for designing antibodies using opponent shaping principles, optimizing antibodies (shapers) against both current and simulated future viral variants. Binding strength is computed via a JAX-accelerated implementation of the Absolut! framework, which relies on discretized binding poses and the Miyazawa–Jernigan energy potential. Viral escape is simulated with an evolutionary algorithm over a horizon of mutations.

While the paper demonstrates that long-horizon shaper antibodies can constrain viral escape more effectively than myopic designs within this simplified simulation, the authors emphasize that the binding and evolution components are interchangeable and acknowledge limitations of the Absolut! binding model and static antigen structure assumptions. Consequently, whether the observed shaping behaviors and performance advantages generalize to more realistic binding simulators and viral evolution models remains unresolved.