Towards scientific discovery with dictionary learning: Extracting biological concepts from microscopy foundation models

2024-12-20

arXiv.org

AI Summary

This study addresses the interpretability bottleneck in microscopy foundation models arising from the lack of biological prior knowledge, proposing a novel paradigm for unsupervised discovery of unknown biological concepts—such as novel cell types or gene perturbation phenotypes—from multi-cell image data. To this end, we introduce Iterative Codebook Feature Learning (ICFL), a method that integrates control-group PCA whitening preprocessing to enable, for the first time on unlabeled biological images, highly selective and interpretable implicit concept extraction. Compared to mainstream sparse autoencoder baselines (e.g., TopK), ICFL substantially improves the biological relevance and specificity of learned features. Our work demonstrates dictionary learning’s potential as a “black-box model decoder” for scientific discovery, establishing a reproducible and generalizable computational framework for interpretability-driven biological knowledge mining with AI.

Technology Category

• Machine Learning: Representation Learning
• Computer Vision: Medical and Biological Imaging
• Cognitive Modeling & Cognitive Systems: Neural Spike Coding

Application Category

• Web Mining and Content Analysis: Large pretrained models with web data
• User Modeling, Personalization and Recommendation: Explainable and interpretable methods for personalization
• Graph Algorithms and Modeling for the Web: Representation, reconstruction, and subgraph or motif discovery in Web-related graphs

Abstract

Dictionary learning (DL) has emerged as a powerful interpretability tool for large language models. By extracting known concepts (e.g., Golden-Gate Bridge) from human-interpretable data (e.g., text), sparse DL can elucidate a model's inner workings. In this work, we ask if DL can also be used to discover unknown concepts from less human-interpretable scientific data (e.g., cell images), ultimately enabling modern approaches to scientific discovery. As a first step, we use DL algorithms to study microscopy foundation models trained on multi-cell image data, where little prior knowledge exists regarding which high-level concepts should arise. We show that sparse dictionaries indeed extract biologically-meaningful concepts such as cell type and genetic perturbation type. We also propose Iterative Codebook Feature Learning~(ICFL) and combine it with a pre-processing step which uses PCA whitening from a control dataset. In our experiments, we demonstrate that both ICFL and PCA improve the selectivity of extracted features compared to TopK sparse autoencoders.

Problem

• Extracts biological concepts from cell images
• Uses dictionary learning for scientific discovery
• Improves feature selectivity with ICFL and PCA

Innovation

• Dictionary learning for concept extraction
• Iterative Codebook Feature Learning
• PCA whitening pre-processing step