A deep learning model to predict RNA-Seq expression of tumours from whole slide images
Problem
Predicting tumour RNA-Seq expression and molecular features from histology whole-slide images without expert annotation or sequencing.
Why it matters
Could enable molecular profiling from routine pathology slides, aiding cancer diagnosis and treatment decisions.
What the AI did
HE2RNA, a deep learning model integrating multiple data modes, was trained to predict RNA-Seq profiles from whole-slide images alone and to spatialize gene expression.
Results
Predicted RNA-Seq profiles from images alone; virtual spatialization validated by CD3- and CD20-staining on an independent dataset; transferable to small datasets.
Impact
Independent staining validation supports the claimed capability, though no numeric performance metrics are given in the abstract.
Record
| Venue | Nature Communications |
|---|---|
| Published | 2020 |
| Work type | article |
| Authors | Benoît Schmauch, Alberto Romagnoni, Elodie Pronier, Charlie Saillard, Pascale Maillé, Julien Caldéraro, Aurélie Kamoun, Meriem Sefta et al. |
| Citations | 525 (OpenAlex, at import time) |
| References | 81 |
| Model or method | HE2RNA |
| Organization | not stated in source |
| Primary source | https://doi.org/10.1038/s41467-020-17678-4 |
| Open access | gold |
| Keywords | deep learning, whole slide images, RNA-Seq, gene expression, cancer, pathology |
Importance rating
significant — thousands to millions affected
Abstract states digital pathology addresses clinical questions from diagnosis to treatment outcomes; predicting molecular features from routine slides could aid cancer diagnosis and treatment decisions.