Mechanistic Interpretability of Antibody Language Models Using SAEs
This advances mechanistic interpretability for domain-specific protein language models, offering insights for researchers in computational biology and AI.
The study tackled the problem of interpreting and controlling antibody language models by applying sparse autoencoders (SAEs), showing that TopK SAEs reveal biologically meaningful features but lack causal control, while Ordered SAEs enable reliable steering at the cost of interpretability.
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 an autoregressive antibody language model, p-IgGen, and steer its 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 an 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 are sufficient for mapping latent features to concepts, Ordered SAEs are preferable when precise generative steering is required.