Related Experiment Video
Updated: May 9, 2025
06:52
Discovery of Driver Genes in Colorectal HT29-derived Cancer Stem-Like Tumorspheres
Published on: July 22, 2020
6.5K
Histology-Based Virtual RNA Inference Identifies Pathways Associated with Metastasis Risk in Colorectal Cancer
Gokul Srinivasan1, Minh-Khang Le1, Zarif Azher1,2
1Departments of Pathology and Laboratory Medicine and Computational Biomedicine, Cedars-Sinai Medical Center, Los Angeles, CA 90048.
Summary
Virtual RNA Inference (VRI) uses standard H&E images to predict spatial transcriptomics data for colorectal cancer (CRC) research. This cost-effective method enhances tumor microenvironment analysis and identifies prognostic gene signatures for better patient outcomes.
Area of Science:
- Computational pathology and bioinformatics
- Molecular epidemiology of cancer
Background:
- Colorectal cancer (CRC) poses a significant health burden, necessitating advancements in screening, prognostication, and treatment.
- The tumor microenvironment (TME) is crucial for CRC progression and metastasis, but traditional characterization methods lack spatial resolution.
- Current spatial transcriptomics (ST) technologies are costly and have limitations for large-scale studies.
Purpose of the Study:
- To refine and implement Virtual RNA Inference (VRI) for deriving ST-level molecular information from H&E-stained tissue images.
- To assess VRI's accuracy in predicting gene expression and estimating cell-type proportions compared to direct ST.
- To identify VRI-derived gene signatures associated with prognostic outcomes in CRC.
Main Methods:
- Trained VRI models on a large CRC ST dataset (45 patients, >300,000 Visium spots) using advanced architectures (UNI, ResNet-50, ViT, VMamba).
- Validated VRI-derived gene signatures against direct ST signatures for various tissue regions.
- Assessed VRI's ability to estimate spatial cell-type proportions from H&E slides.
- Analyzed VRI-derived gene signatures for association with prognostic outcomes in an expanded CRC cohort.
Main Results:
- Achieved a median Spearman's correlation of 0.546 between predicted and measured spot-level gene expression.
- Demonstrated strong concordance between VRI-derived and direct ST gene signatures for tissue regions.
- Showed accurate spatial estimation of cell-type proportions from H&E slides.
- Identified VRI-derived gene signatures significantly associated with metastasis status in CRC patients.
Conclusions:
- VRI effectively infers ST-level molecular information from standard H&E images, offering a cost-effective alternative to direct ST.
- VRI enables near-cellular resolution analysis of the TME and identifies clinically relevant prognostic gene signatures.
- This approach has the potential to accelerate large-scale translational CRC research by leveraging existing H&E-stained tissue archives.
