Predicting Spatially Resolved Gene Expression from Histopathology Images
Key Highlights
A biopharma company partnered with Elucidata and we developed an AI model to predict spatially resolved gene expression directly from H&E-stained histopathology images, reducing the need for costly spatial transcriptomics experiments.
The solution reduces experimental costs by 40% and speeds up biomarker analysis from months to weeks (4X), making it more scalable and cost-effective.
The model leverages a CLIP-like contrastive learning framework, multi-scale feature extraction, and cell type-based inference for accurate gene expression prediction.
The model achieved a 0.41 Spearman's correlation for measured genes, significantly outperforming traditional methods and successfully predicting out-of-distribution genes.
Thank you for showing interest! You will receive the Case Study in your Mailbox shortly.
.svg)
.svg)
%201%20(1).svg)
.svg)
%201.svg){kind=icon}
.svg)
%20(1).svg)
