Discovering Latent Scientific Concepts through Discrete Representation Learning
- 1MSKCC
Abstract
Scientific imaging is undergoing a fundamental transformation. Advances in multiplex immunofluorescence, computational pathology, and electron microscopy now enable observation of biological and physical systems at unprecedented spatial resolution. Rather than simply classifying known objects, these data offer the opportunity to discover previously unknown organizational principles. However, scientific image datasets are typically highly curated, information-rich, and limited in size, making reliable representation learning particularly challenging. We show that standard continuous self-supervised adaptation can become unstable in this regime, allowing teacher representations to drift and degrading downstream transfer. We introduce Kodiak, a discrete representation learning architecture that discovers latent scientific concepts through Sinkhorn-balanced codebook assignments. Instead of learning unconstrained continuous embeddings, Kodiak constructs a reusable vocabulary of discrete concepts that provides stable supervision for both masked-patch prediction and cross-view representation learning within a single-stage framework. Across five scientific imaging domains, including multiplex immunofluorescence for pancreatic cancer pathology and scanning electron microscopy, Kodiak consistently improves representation quality under both continued foundation-model adaptation and training from scratch. Beyond improved classification and transfer,the learned concepts are interpretable, reusable, and produce spatially coherent concept maps that support applications such as cell phenotyping, segmentation-free patch phenotyping, spatial niche discovery, tissue organization, and materials characterization. We view Kodiak as the representation-learning engine of our foundation model for spatial biology. More broadly, our results suggest that discrete representation learning provides a principled foundation for discovering reusable latent scientific concepts that can accelerate scientific interpretation across imaging domains.
Related
- SYNERGY-VAE: An explainable generative deep learning framework for discovering depression subgroups from multimodal population health data
- Learning general representation of 12-lead ECG with a joint-embedding predictive architecture
- An Explainable Disease-Aware Intelligent Histopathological Representation Learning Framework for Generalizable Breast Cancer Classification
- From Learning Theory to Fragmentation: CBT and the Problem of Borrowed Scientific Credibility
- Digital transformation in higher scientific education: The roles of virtual simulation, AI-assisted learning, and blended teaching
BibTeX
@misc{stoimcev2026discovering,
title = {{Discovering Latent Scientific Concepts through Discrete Representation Learning}},
author = {M. Stoimcev and M. Verma and A. Filliol and M. Skamagki and S. Flowers and J. Cohen and Z. Azadian and N. Newlin and M. S. Abdel-Mottaleb and S. Dzeroski and P. Romesser and S. Lowe and N. Dimitrova},
howpublished = {bioRxiv},
year = {2026},
month = oct,
doi = {10.64898/2026.10.03.756431},
url = {https://www.biorxiv.org/content/10.64898/2026.10.03.756431v1},
note = {CC BY-NC-ND},
}
RIS
TY - UNPB
TI - Discovering Latent Scientific Concepts through Discrete Representation Learning
AU - M. Stoimcev
AU - M. Verma
AU - A. Filliol
AU - M. Skamagki
AU - S. Flowers
AU - J. Cohen
AU - Z. Azadian
AU - N. Newlin
AU - M. S. Abdel-Mottaleb
AU - S. Dzeroski
AU - P. Romesser
AU - S. Lowe
AU - N. Dimitrova
PY - 2026
DA - 2026/10/03
PB - bioRxiv
DO - 10.64898/2026.10.03.756431
UR - https://www.biorxiv.org/content/10.64898/2026.10.03.756431v1
AB - Scientific imaging is undergoing a fundamental transformation. Advances in multiplex immunofluorescence, computational pathology, and electron microscopy now enable observation of biological and physical systems at unprecedented spatial resolution. Rather than simply classifying known objects, these data offer the opportunity to discover previously unknown organizational principles. However, scientific image datasets are typically highly curated, information-rich, and limited in size, making reliable representation learning particularly challenging. We show that standard continuous self-supervised adaptation can become unstable in this regime, allowing teacher representations to drift and degrading downstream transfer. We introduce Kodiak, a discrete representation learning architecture that discovers latent scientific concepts through Sinkhorn-balanced codebook assignments. Instead of learning unconstrained continuous embeddings, Kodiak constructs a reusable vocabulary of discrete concepts that provides stable supervision for both masked-patch prediction and cross-view representation learning within a single-stage framework. Across five scientific imaging domains, including multiplex immunofluorescence for pancreatic cancer pathology and scanning electron microscopy, Kodiak consistently improves representation quality under both continued foundation-model adaptation and training from scratch. Beyond improved classification and transfer,the learned concepts are interpretable, reusable, and produce spatially coherent concept maps that support applications such as cell phenotyping, segmentation-free patch phenotyping, spatial niche discovery, tissue organization, and materials characterization. We view Kodiak as the representation-learning engine of our foundation model for spatial biology. More broadly, our results suggest that discrete representation learning provides a principled foundation for discovering reusable latent scientific concepts that can accelerate scientific interpretation across imaging domains.
ER -