Unveiling Latent Knowledge in Chemistry Language Models through Sparse Autoencoders

• URL: http://arxiv.org/abs/2512.08077v1
• Date: Mon, 08 Dec 2025 22:20:01 GMT
• Title: Unveiling Latent Knowledge in Chemistry Language Models through Sparse Autoencoders
• Authors: Jaron Cohen, Alexander G. Hasson, Sara Tanovic,
• Abstract summary: We extend sparse autoencoder techniques to uncover and examine interpretable features within chemistry language models.<n>Our findings reveal that these models encode a rich landscape of chemical concepts.<n>Our approach provides a generalisable framework for uncovering latent knowledge in chemistry-focused AI systems.
• Score: 42.033443425253644
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Since the advent of machine learning, interpretability has remained a persistent challenge, becoming increasingly urgent as generative models support high-stakes applications in drug and material discovery. Recent advances in large language model (LLM) architectures have yielded chemistry language models (CLMs) with impressive capabilities in molecular property prediction and molecular generation. However, how these models internally represent chemical knowledge remains poorly understood. In this work, we extend sparse autoencoder techniques to uncover and examine interpretable features within CLMs. Applying our methodology to the Foundation Models for Materials (FM4M) SMI-TED chemistry foundation model, we extract semantically meaningful latent features and analyse their activation patterns across diverse molecular datasets. Our findings reveal that these models encode a rich landscape of chemical concepts. We identify correlations between specific latent features and distinct domains of chemical knowledge, including structural motifs, physicochemical properties, and pharmacological drug classes. Our approach provides a generalisable framework for uncovering latent knowledge in chemistry-focused AI systems. This work has implications for both foundational understanding and practical deployment; with the potential to accelerate computational chemistry research.

Related papers

• Agentic reinforcement learning empowers next-generation chemical language models for molecular design and synthesis [51.83339196548892]
  ChemCraft is a novel framework that decouples chemical reasoning from knowledge storage.<n>ChemCraft achieves superior performance with minimal inference costs.<n>This work establishes a cost-effective and privacy-preserving paradigm for AI-aided chemistry.
  arXiv  Detail & Related papers  (2026-01-25T04:23:34Z)
• Foundation Models for Discovery and Exploration in Chemical Space [57.97784111110166]
  MIST is a family of molecular foundation models trained on large unlabeled datasets.<n>We demonstrate the ability of these models to solve real-world problems across chemical space.
  arXiv  Detail & Related papers  (2025-10-20T17:56:01Z)
• $\text{M}^{2}$LLM: Multi-view Molecular Representation Learning with Large Language Models [59.125833618091846]
  We propose a multi-view framework that integrates three perspectives: the molecular structure view, the molecular task view, and the molecular rules view.<n>Experiments demonstrate that $textM2$LLM achieves state-of-the-art performance on multiple benchmarks across classification and regression tasks.
  arXiv  Detail & Related papers  (2025-08-12T05:46:47Z)
• Knowledge-aware contrastive heterogeneous molecular graph learning [77.94721384862699]
  We propose a paradigm shift by encoding molecular graphs into Heterogeneous Molecular Graph Learning (KCHML)<n>KCHML conceptualizes molecules through three distinct graph views-molecular, elemental, and pharmacological-enhanced by heterogeneous molecular graphs and a dual message-passing mechanism.<n>This design offers a comprehensive representation for property prediction, as well as for downstream tasks such as drug-drug interaction (DDI) prediction.
  arXiv  Detail & Related papers  (2025-02-17T11:53:58Z)
• Leveraging Chemistry Foundation Models to Facilitate Structure Focused Retrieval Augmented Generation in Multi-Agent Workflows for Catalyst and Materials Design [0.0]
  We show that chemistry foundation models can serve as a basis for enabling structure-focused, semantic chemistry information retrieval.<n>We also show the use of chemistry foundation models in conjunction with multi-modal models such as OpenCLIP.
  arXiv  Detail & Related papers  (2024-08-21T17:25:45Z)
• MolTRES: Improving Chemical Language Representation Learning for Molecular Property Prediction [14.353313239109337]
  MolTRES is a novel chemical language representation learning framework. It incorporates generator-discriminator training, allowing the model to learn from more challenging examples. Our model outperforms existing state-of-the-art models on popular molecular property prediction tasks.
  arXiv  Detail & Related papers  (2024-07-09T01:14:28Z)
• Multi-channel learning for integrating structural hierarchies into context-dependent molecular representation [10.025809630976065]
  This paper introduces a novel pre-training framework that learns robust and generalizable chemical knowledge.<n>Our approach demonstrates competitive performance across various molecular property benchmarks.
  arXiv  Detail & Related papers  (2023-11-05T23:47:52Z)
• Unsupervised Learning of Molecular Embeddings for Enhanced Clustering and Emergent Properties for Chemical Compounds [2.6803933204362336]
  We introduce various methods to detect and cluster chemical compounds based on their SMILES data. Our first method, analyzing the graphical structures of chemical compounds using embedding data, employs vector search to meet our threshold value. We also used natural language description embeddings stored in a vector database with GPT3.5, which outperforms the base model.
  arXiv  Detail & Related papers  (2023-10-25T18:00:24Z)
• Structure to Property: Chemical Element Embeddings and a Deep Learning Approach for Accurate Prediction of Chemical Properties [0.0]
  We introduce the elEmBERT model for chemical classification tasks. It is based on deep learning techniques, such as a multilayer encoder architecture. We demonstrate the opportunities offered by our approach on sets of organic, inorganic and crystalline compounds.
  arXiv  Detail & Related papers  (2023-09-17T19:41:32Z)
• Implicit Geometry and Interaction Embeddings Improve Few-Shot Molecular Property Prediction [53.06671763877109]
  We develop molecular embeddings that encode complex molecular characteristics to improve the performance of few-shot molecular property prediction. Our approach leverages large amounts of synthetic data, namely the results of molecular docking calculations. On multiple molecular property prediction benchmarks, training from the embedding space substantially improves Multi-Task, MAML, and Prototypical Network few-shot learning performance.
  arXiv  Detail & Related papers  (2023-02-04T01:32:40Z)
• A Molecular Multimodal Foundation Model Associating Molecule Graphs with Natural Language [63.60376252491507]
  We propose a molecular multimodal foundation model which is pretrained from molecular graphs and their semantically related textual data. We believe that our model would have a broad impact on AI-empowered fields across disciplines such as biology, chemistry, materials, environment, and medicine.
  arXiv  Detail & Related papers  (2022-09-12T00:56:57Z)

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.