Mechanistic Interpretability of Antibody Language Models Using SAEs
Abstract
Sparse autoencoders (SAEs) are a mechanistic interpretability technique that have been used to provide insight into learned concepts within large protein language models. Here, we employ TopK and Ordered SAEs to investigate autoregressive antibody language models, and steer their generation. We show that TopK SAEs can reveal biologically meaningful latent features, but high feature-concept correlation does not guarantee causal control over generation. In contrast, Ordered SAEs impose a hierarchical structure that reliably identifies steerable features, but at the expense of more complex and less interpretable activation patterns. These findings advance the mechanistic interpretability of domain-specific protein language models and suggest that, while TopK SAEs suffice for mapping latent features to concepts, Ordered SAEs are preferable when precise generative steering is required.
Cite
@article{arxiv.2512.05794,
title = {Mechanistic Interpretability of Antibody Language Models Using SAEs},
author = {Rebonto Haque and Oliver M. Turnbull and Anisha Parsan and Nithin Parsan and John J. Yang and Anna L. Beukenhorst and Charlotte M. Deane},
journal= {arXiv preprint arXiv:2512.05794},
year = {2026}
}
Comments
v3: 15 pages; corrected author list and affiliations in the main text; minor text changes; updated steering results following minor code changes; conclusions and findings remain unchanged; included link to data and code in the Data Availability section