Barman, R. K. | bioRxiv Science

bioRxiv2025

Pathologist-interpretable breast cancer subtyping and stratification from AI-inferred nuclear features

Artificial intelligence (AI) is making notable advances in digital pathology but faces challenges in human interpretability. Here we introduce EXPAND (EXplainable Pathologist Aligned Nuclear Discriminator), the first pathologist-interpretable AI model to predict breast cancer tumor subtypes and patient survival. EXPAND focuses on a core set of 12 nuclear pathologist-interpretable features (NPIFs), composing the Nottingham grading criteria used by the pathologists. It is a fully automated, end-to-end diagnostic workflow, which automatically extracts NPIFs given a patient tumor slide and uses them to predict tumor subtype and survival. EXPANDs performance is comparable to that of existing deep learning non-interpretable black box AI models. It achieves areas under the ROC curve (AUC) values of 0.73, 0.79 and 0.75 for predicting HER2+, HR+ and TNBC tumor subtypes, respectively, matching the performance of proprietary models that rely on substantially larger and more complex interpretable feature sets. The 12 NPIFs demonstrate strong and independent prognostic value for patient survival, underscoring their potential as biologically grounded, interpretable biomarkers for survival stratification in BC. These results lay the basis for building interpretable AI diagnostic models in other cancer indications. The complete end-to-end pipeline is made publicly available via GitHub (https://github.com/ruppinlab/EXPAND) to support community use and reproducibility.

bioinformatics↗

bioRxiv2025

Path2Omics: Enhanced transcriptomic and methylation prediction accuracy from tumor histopathology

Precision oncology is becoming increasingly integral to clinical practice, demonstrating notable improvements in treatment outcomes. While molecular data provide comprehensive insights, obtaining such data remains costly and time-consuming. To address this challenge, we developed Path2Omics, a deep learning model that predicts gene expression and methylation from histopathology for 23 cancer types. Path2Omics was trained on 20,497 slides (9,456 formalin-fixed and paraffin-embedded (FFPE) and 11,041 fresh frozen (FF)) from 8,007 patients across 23 The Cancer Genome Atlas cohorts. When tested on FFPE slides, the most readily available format in clinical pathology practice, the integrated model outperformed its individual FF and FFPE components, robustly predicting nearly 5,000 genes on average, approximately five times more than our recently published DeepPT model. Externally evaluated on seven independent cohorts, Path2Omics robustly predicted the expression of approximately 4,400 genes, yielding a 30% increase over the FFPE model alone. Finally, we demonstrate that the inferred gene expression is nearly as effective as the actual values in predicting patient survival and treatment response. These results lay the basis for using Path2Omics to advance precision oncology from histopathology slides in a speedy and cost-effective manner. Statement of significancePath2Omics is a deep learning model that accurately predicts gene expression and methylation from histopathology slides across 23 cancer types. Unlike existing approaches that rely solely on FFPE slides for training, Path2Omics leverages both FFPE and FF slides by constructing two separate models and integrating them. Downstream analyses show that the inferred values from Path2Omics are nearly as effective as actual values in predicting patient survival and treatment response.

bioinformatics↗