Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

HistoPrism: Unlocking Functional Pathway Analysis from Pan-Cancer Histology via Gene Expression Prediction

Susu Hu, Qinghe Zeng, Nithya Bhasker, Jakob Nicolas Kather, Stefanie Speidel

TL;DR

HistoPrism introduces a pan-cancer transformer that directly maps histology-derived patch features to gene expression across cancer types, using pan-cancer conditioning and a Transformer encoder to capture contextual tissue structure. It pairs this architecture with Gene Pathway Coherence (GPC), a benchmark based on Hallmark and Gene Ontology pathways, to evaluate biological coherence rather than relying solely on variance-based metrics. The approach achieves state-of-the-art performance on highly variable genes and demonstrates robust pathway-level coherence, improved clustering across ~38k genes, and substantially better data and computational efficiency than prior methods. Collectively, these advances move histology-to-transcriptomics toward clinically scalable deployment by emphasizing functional coherence and efficient, cross-cancer generalization.

Abstract

Predicting spatial gene expression from H&E histology offers a scalable and clinically accessible alternative to sequencing, but realizing clinical impact requires models that generalize across cancer types and capture biologically coherent signals. Prior work is often limited to per-cancer settings and variance-based evaluation, leaving functional relevance underexplored. We introduce HistoPrism, an efficient transformer-based architecture for pan-cancer prediction of gene expression from histology. To evaluate biological meaning, we introduce a pathway-level benchmark, shifting assessment from isolated gene-level variance to coherent functional pathways. HistoPrism not only surpasses prior state-of-the-art models on highly variable genes , but also more importantly, achieves substantial gains on pathway-level prediction, demonstrating its ability to recover biologically coherent transcriptomic patterns. With strong pan-cancer generalization and improved efficiency, HistoPrism establishes a new standard for clinically relevant transcriptomic modeling from routinely available histology.

HistoPrism: Unlocking Functional Pathway Analysis from Pan-Cancer Histology via Gene Expression Prediction

TL;DR

HistoPrism introduces a pan-cancer transformer that directly maps histology-derived patch features to gene expression across cancer types, using pan-cancer conditioning and a Transformer encoder to capture contextual tissue structure. It pairs this architecture with Gene Pathway Coherence (GPC), a benchmark based on Hallmark and Gene Ontology pathways, to evaluate biological coherence rather than relying solely on variance-based metrics. The approach achieves state-of-the-art performance on highly variable genes and demonstrates robust pathway-level coherence, improved clustering across ~38k genes, and substantially better data and computational efficiency than prior methods. Collectively, these advances move histology-to-transcriptomics toward clinically scalable deployment by emphasizing functional coherence and efficient, cross-cancer generalization.

Abstract

Predicting spatial gene expression from H&E histology offers a scalable and clinically accessible alternative to sequencing, but realizing clinical impact requires models that generalize across cancer types and capture biologically coherent signals. Prior work is often limited to per-cancer settings and variance-based evaluation, leaving functional relevance underexplored. We introduce HistoPrism, an efficient transformer-based architecture for pan-cancer prediction of gene expression from histology. To evaluate biological meaning, we introduce a pathway-level benchmark, shifting assessment from isolated gene-level variance to coherent functional pathways. HistoPrism not only surpasses prior state-of-the-art models on highly variable genes , but also more importantly, achieves substantial gains on pathway-level prediction, demonstrating its ability to recover biologically coherent transcriptomic patterns. With strong pan-cancer generalization and improved efficiency, HistoPrism establishes a new standard for clinically relevant transcriptomic modeling from routinely available histology.
Paper Structure (28 sections, 5 equations, 5 figures, 9 tables)

This paper contains 28 sections, 5 equations, 5 figures, 9 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Computational Prediction of Spatial Transcriptomics
  4. Foundation Models in Digital Pathology
  5. Methodology
  6. Problem Formulation
  7. HistoPrism: A Direct-Mapping Architecture
  8. 1. Pan-Cancer Conditioning via Cross-Attention.
  9. 2. Contextual Aggregation with a Transformer Encoder.
  10. 3. Gene Expression Regression.
  11. A Framework for Evaluating Biological Coherence
  12. Baseline Metric: Highly Variant Gene Correlation.
  13. The Gene Pathway Coherence (GPC) Benchmark.
  14. Experiments and Results
  15. Experimental Setup
  16. ...and 13 more sections

Figures (5)

  • Figure 1: HistoPrism architecture. Patch level image embeddings are obtained via pathology foundation models. A cross-attention module injects pan-cancer conditioning. A Transformer Encoder models contextual relations before a final MLP head regresses gene expression values.
  • Figure 2: Comparison of gene pathway coherence (GPC) in PCC on both Hallmark gene pathways and Gene Ontology pathways.
  • Figure 3: Model efficiency comparison of HistoPrism and STPath in terms of forward pass runtime, peak GPU memory usage, and FLOPs across different numbers of patches.
  • Figure 4: Ablation study of the impact of PFM Gigapath on our model HistoPrism's GPC performance.
  • Figure 5: Gene variance distribution density plot.