Related Experiment Video
Updated: Sep 18, 2025
07:34
Author Spotlight: Advancing Prostate Cancer Research Through Improved Tissue Sampling and Biobanking
Published on: November 17, 2023
793
Path2Omics: Enhanced transcriptomic and methylation prediction accuracy from tumor histopathology.
Danh-Tai Hoang1,2, Eldad D Shulman1, Saugato Rahman Dhruba1
1Cancer Data Science Laboratory, Center for Cancer Research, National Cancer Institute, Bethesda, MD, USA.
Biorxiv : the Preprint Server for Biology
|June 26, 2025
Summary
Path2Omics, a deep learning model, predicts gene expression and methylation from histopathology slides for 23 cancer types. This cost-effective approach advances precision oncology by enabling faster, more accessible molecular insights from tissue samples.
Area of Science:
- Computational biology
- Genomics
- Pathology
Background:
- Precision oncology relies on molecular data, but obtaining it is costly and time-consuming.
- Histopathology slides are readily available in clinical settings.
Purpose of the Study:
- To develop Path2Omics, a deep learning model predicting gene expression and methylation from histopathology slides.
- To enable cost-effective and rapid molecular insights for precision oncology.
Main Methods:
- Trained Path2Omics on 20,497 slides (FFPE and FF) from 8,007 patients across 23 TCGA cohorts.
- Integrated separate FFPE and FF models for enhanced prediction accuracy.
- Validated externally on seven independent cohorts.
Main Results:
- Path2Omics robustly predicted nearly 5,000 genes on average from FFPE slides, outperforming previous models.
- Achieved a 30% increase in gene prediction over FFPE-only models on external cohorts.
- Inferred gene expression effectively predicted patient survival and treatment response, comparable to actual values.
Conclusions:
- Path2Omics offers a speedy and cost-effective method to derive molecular insights from histopathology slides.
- This deep learning approach can significantly advance precision oncology by leveraging readily available tissue data.
