ProtSAE: Disentangling and Interpreting Protein Language Models via Semantically-Guided Sparse Autoencoders

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Liu, Xiangyu, Lei, Haodi, Liu, Yi, Liu, Yang, Hu, Wei
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866917208888705024
author Liu, Xiangyu
Lei, Haodi
Liu, Yi
Liu, Yang
Hu, Wei
author_facet Liu, Xiangyu
Lei, Haodi
Liu, Yi
Liu, Yang
Hu, Wei
contents Sparse Autoencoder (SAE) has emerged as a powerful tool for mechanistic interpretability of large language models. Recent works apply SAE to protein language models (PLMs), aiming to extract and analyze biologically meaningful features from their latent spaces. However, SAE suffers from semantic entanglement, where individual neurons often mix multiple nonlinear concepts, making it difficult to reliably interpret or manipulate model behaviors. In this paper, we propose a semantically-guided SAE, called ProtSAE. Unlike existing SAE which requires annotation datasets to filter and interpret activations, we guide semantic disentanglement during training using both annotation datasets and domain knowledge to mitigate the effects of entangled attributes. We design interpretability experiments showing that ProtSAE learns more biologically relevant and interpretable hidden features compared to previous methods. Performance analyses further demonstrate that ProtSAE maintains high reconstruction fidelity while achieving better results in interpretable probing. We also show the potential of ProtSAE in steering PLMs for downstream generation tasks.
format Preprint
id arxiv_https___arxiv_org_abs_2509_05309
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle ProtSAE: Disentangling and Interpreting Protein Language Models via Semantically-Guided Sparse Autoencoders
Liu, Xiangyu
Lei, Haodi
Liu, Yi
Liu, Yang
Hu, Wei
Quantitative Methods
Artificial Intelligence
Computation and Language
Sparse Autoencoder (SAE) has emerged as a powerful tool for mechanistic interpretability of large language models. Recent works apply SAE to protein language models (PLMs), aiming to extract and analyze biologically meaningful features from their latent spaces. However, SAE suffers from semantic entanglement, where individual neurons often mix multiple nonlinear concepts, making it difficult to reliably interpret or manipulate model behaviors. In this paper, we propose a semantically-guided SAE, called ProtSAE. Unlike existing SAE which requires annotation datasets to filter and interpret activations, we guide semantic disentanglement during training using both annotation datasets and domain knowledge to mitigate the effects of entangled attributes. We design interpretability experiments showing that ProtSAE learns more biologically relevant and interpretable hidden features compared to previous methods. Performance analyses further demonstrate that ProtSAE maintains high reconstruction fidelity while achieving better results in interpretable probing. We also show the potential of ProtSAE in steering PLMs for downstream generation tasks.
title ProtSAE: Disentangling and Interpreting Protein Language Models via Semantically-Guided Sparse Autoencoders
topic Quantitative Methods
Artificial Intelligence
Computation and Language
url https://arxiv.org/abs/2509.05309