Interpreting Latent Protein Language Model Features with Geometric Annotations

• URL: http://arxiv.org/abs/2608.26419v1
• Date: Wed, 26 Aug 2026 21:39:50 GMT
• Title: Interpreting Latent Protein Language Model Features with Geometric Annotations
• Abstract summary: Sparse autoencoders (SAEs) provide a promising tool to disentangle latent pLM representations into interpretable features.<n>We introduce an automated and scalable method for interpreting SAE features in ESM-2 by using geometrically inspired features of the protein $textC_$ backbone.<n>A significant portion of SAE features activate on unannotated metagenomic protein sequences enabling us to use our SAE annotations to better understand these sequences.
• Score: 2.6126272668390373
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Protein language models (pLMs) encode information about protein sequences which enable downstream tasks such as structure prediction, but their internal representations are not well understood. Sparse autoencoders (SAEs) provide a promising tool to disentangle latent pLM representations into interpretable features, but existing annotation pipelines largely rely on protein-level annotations derived from database labels and LLM annotations of top activating sequences. Such annotations can overlook the localized residue-level and geometric patterns encoded by sparse features. We introduce an automated and scalable method for interpreting SAE features in ESM-2 by using geometrically inspired features of the protein $\text{C}_α$ backbone. Across ESM-2 8M layers, an FDR-controlled discovery analysis shows that local geometry is significantly associated with many SAE features, with varying levels of predictive strength, expanding coverage beyond database and sequence-based methods. In particular, geometry can distinguish SAE features sharing the same database annotation, revealing substructure within known biological labels. A significant portion of SAE features activate on unannotated metagenomic protein sequences enabling us to use our SAE annotations to better understand these sequences. In addition, ablation experiments at the level of contact prediction show that removing found geometric features shifts ESM-2's predicted contact maps in the direction of the descriptor. This provides a robust method of annotating proteins activated within SAE neurons at a residue level, providing a bridge between mechanistic interpretability and structural biology.

Related papers

• From Sparse Features to Trustworthy Proxies: Certifying SAE-Based Interpretability [53.681678236115836]
  We study when can an SAE-based explanation be treated as a faithful view of an underlying frozen LM.<n>Our framework derives an upper bound on the base model's expected risk using four measurable quantities.<n>We show that the bound becomes non-vacuous on GPT-2 Small, Gemma-2B, and Llama-3-8B at practical sample sizes.
  arXiv  Detail & Related papers  (2026-06-16T18:28:23Z)
• From Atoms to Trees: Building a Structured Feature Forest with Hierarchical Sparse Autoencoders [12.294297654425334]
  Hierarchical Sparse Autoencoder (HSAE) learns a series of SAEs and the parent-child relationships between their features.<n>Our work provides a powerful, scalable tool for discovering and analyzing the multi-scale conceptual structures embedded in language representations.
  arXiv  Detail & Related papers  (2026-02-12T12:30:23Z)
• S$^2$Drug: Bridging Protein Sequence and 3D Structure in Contrastive Representation Learning for Virtual Screening [72.89086338778098]
  We propose a two-stage framework for protein-ligand contrastive representation learning.<n>In the first stage, we perform protein sequence pretraining on ChemBL using an ESM2-based backbone.<n>In the second stage, we fine-tune on PDBBind by fusing sequence and structure information through a residue-level gating module.<n>This auxiliary task guides the model to accurately localize binding residues within the protein sequence and capture their 3D spatial arrangement.
  arXiv  Detail & Related papers  (2025-11-10T11:57:47Z)
• Protein as a Second Language for LLMs [50.34983283157322]
  "Protein-as-Second-Language" framework reformulates amino-acid sequences as sentences in a novel symbolic language.<n>We curate a bilingual corpus of 79,926 protein-QA instances spanning attribute prediction, descriptive understanding, and extended reasoning.<n>Our method delivers consistent gains across diverse open-source LLMs and GPT-4, achieving up to 17.2% ROUGE-L improvement.
  arXiv  Detail & Related papers  (2025-10-13T09:21:45Z)
• Towards Understanding the Shape of Representations in Protein Language Models [0.0]
  We identify protein structure and representation with square-root velocity (SRV) representations and graph filtrations.<n>We find that PLMs preferentially encode immediate as well as local relations between residues, but start to degrade for larger context lengths.
  arXiv  Detail & Related papers  (2025-09-29T15:06:24Z)
• ProtSAE: Disentangling and Interpreting Protein Language Models via Semantically-Guided Sparse Autoencoders [30.219733023958188]
  Sparse Autoencoder (SAE) has emerged as a powerful tool for mechanistic interpretability of large language models.<n>We propose a semantically-guided SAE, called ProtSAE.<n>We show that ProtSAE learns more biologically relevant and interpretable hidden features compared to previous methods.
  arXiv  Detail & Related papers  (2025-08-26T11:20:31Z)
• Explainable Mapper: Charting LLM Embedding Spaces Using Perturbation-Based Explanation and Verification Agents [11.168089496463125]
  Large language models (LLMs) produce high-dimensional embeddings that capture rich semantic and syntactic relationships between words, sentences, and concepts.<n>We introduce a framework for semi-automatic annotation of these embedding properties.
  arXiv  Detail & Related papers  (2025-07-24T17:43:40Z)
• PRING: Rethinking Protein-Protein Interaction Prediction from Pairs to Graphs [88.98041407783502]
  PRING is the first benchmark that evaluates protein-protein interaction prediction from a graph-level perspective.<n> PRING curates a high-quality, multi-species PPI network dataset comprising 21,484 proteins and 186,818 interactions.
  arXiv  Detail & Related papers  (2025-07-07T15:21:05Z)
• Dense SAE Latents Are Features, Not Bugs [86.50389855919292]
  We show that dense latents serve functional roles in language model computation.<n>We identify classes tied to position tracking, context binding, entropy regulation, letter-specific output signals, part-of-speech, and principal component reconstruction.
  arXiv  Detail & Related papers  (2025-06-18T17:59:35Z)
• Towards Interpretable Protein Structure Prediction with Sparse Autoencoders [0.0]
  Matryoshka SAEs learn hierarchically organized features by forcing nested groups of latents to reconstruct inputs independently.<n>We scale SAEs to ESM2-3B, the base model for ESMFold, enabling mechanistic interpretability of protein structure prediction for the first time.<n>We show that SAEs trained on ESM2-3B significantly outperform those trained on smaller models for both biological concept discovery and contact map prediction.
  arXiv  Detail & Related papers  (2025-03-11T17:57:29Z)
• InterPLM: Discovering Interpretable Features in Protein Language Models via Sparse Autoencoders [24.150250149027883]
  Protein language models (PLMs) have demonstrated remarkable success in protein modeling and design.<n>We present a systematic approach to extract and analyze interpretable features from PLMs using sparse autoencoders.<n>As practical applications, we demonstrate how these latent features can fill in missing annotations in protein databases.
  arXiv  Detail & Related papers  (2024-11-13T18:51:21Z)
• Llama Scope: Extracting Millions of Features from Llama-3.1-8B with Sparse Autoencoders [115.34050914216665]
  Sparse Autoencoders (SAEs) have emerged as a powerful unsupervised method for extracting sparse representations from language models. We introduce a suite of 256 SAEs, trained on each layer and sublayer of the Llama-3.1-8B-Base model, with 32K and 128K features. We assess the generalizability of SAEs trained on base models to longer contexts and fine-tuned models.
  arXiv  Detail & Related papers  (2024-10-27T17:33:49Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.