Learning to predict RNA sequence expressions from whole slide images with applications for search and classification
Problem
Molecular tests like bulk RNA-seq are expensive, time-consuming and require additional tissue, limiting their use alongside pathology slide images.
Why it matters
Avoiding costly, slow molecular tests could make tissue-based molecular information more accessible.
What the AI did
tRNAsformer, an attention-based model using multiple instance learning, learns to predict bulk RNA-seq from whole slide images and to represent slides for search and classification.
Results
Better performance and faster convergence compared to state-of-the-art algorithms; no numbers stated.
Impact
Authors frame it as a computational pathology tool to assist search and classification; clinical impact not shown.
Record
| Venue | Communications Biology |
|---|---|
| Published | 2023 |
| Work type | article |
| Authors | Areej Alsaafin, Amir Safarpoor, Milad Sikaroudi, Jason Hipp, Hamid R. Tizhoosh |
| Citations | 40 (OpenAlex, at import time) |
| References | 33 |
| Model or method | tRNAsformer |
| Organization | not stated in source |
| Primary source | https://doi.org/10.1038/s42003-023-04583-x |
| Open access | gold |
| Keywords | whole slide images, RNA-seq prediction, multiple instance learning, computational pathology, attention |
Importance rating
significant — thousands to millions affected
Avoiding expensive, slow molecular tests could widen access to tissue-based molecular information in pathology; a genuinely new capability but clinical impact not shown.