Protein Circuit Tracing via Cross-layer Transcoders
This addresses the challenge of interpretability in protein language models for researchers in computational biology, offering a novel method to trace circuits with practical applications in protein design.
The paper tackled the problem of understanding computational circuits in protein language models by introducing ProtoMech, a framework using cross-layer transcoders, which recovered 82-89% of original performance on tasks and enabled high-fitness protein design surpassing baselines in over 70% of cases.
Protein language models (pLMs) have emerged as powerful predictors of protein structure and function. However, the computational circuits underlying their predictions remain poorly understood. Recent mechanistic interpretability methods decompose pLM representations into interpretable features, but they treat each layer independently and thus fail to capture cross-layer computation, limiting their ability to approximate the full model. We introduce ProtoMech, a framework for discovering computational circuits in pLMs using cross-layer transcoders that learn sparse latent representations jointly across layers to capture the model's full computational circuitry. Applied to the pLM ESM2, ProtoMech recovers 82-89% of the original performance on protein family classification and function prediction tasks. ProtoMech then identifies compressed circuits that use <1% of the latent space while retaining up to 79% of model accuracy, revealing correspondence with structural and functional motifs, including binding, signaling, and stability. Steering along these circuits enables high-fitness protein design, surpassing baseline methods in more than 70% of cases. These results establish ProtoMech as a principled framework for protein circuit tracing.