Simulating protein evolution with programmable fitness objectives derived from modern protein models

Programmable fitness objectives derived from protein models can simulate amino-acid substitution evolution that reproduces distinct natural sequence constraints, including interface-specific effects in homodimers.

• Why it matters: Classical simulations use fixed substitution rules or single-proxy fitness, limiting how well they capture the complex, constraint-driven nature of selection in natural proteins.
• What they did: The study simulated protein evolution under programmable fitness objectives from protein language models, inverse folding, physics-based energy functions, and alignment-based statistical models, including combined objectives and oligomeric or multi-conformation structural contexts.
• The result: Different fitness proxies created distinct realism–plausibility trade-offs, and backpropagating structural objectives through AlphaFold2 produced evolved sequences with favorable molecular-property shifts versus discrete stepwise mutation-selection while retaining key evolutionary signatures.