Genome-anchored foundation model embeddings improve molecular prediction from histology images

Precision oncology requires accurate molecular insights, yet obtaining these directly from genomics is costly and time-consuming for broad clinical use. Predicting complex molecular features and patient prognosis directly from routine whole-slide images (WSIs) remains a major challenge for current deep learning methods. Here we introduce PathLUPI, which uses transcriptomic privileged information during training to extract genome-anchored histological embeddings, enabling effective molecular prediction using only WSIs at inference. Through extensive evaluation across 49 molecular oncology tasks using 11,361 cases from 20 cohorts, PathLUPI demonstrates superior performance compared to conventional methods trained solely on WSIs. Crucially, it achieves an area under the receiver operating characteristic curve of at least 0.80 across 18 biomarker prediction and molecular subtyping tasks, and a concordance index of at least 0.70 across survival cohorts spanning five major cancer types. Moreover, PathLUPI embeddings reveal distinct cellular morphological signatures associated with specific genotypes, whose spatial fidelity is substantiated in representative high-resolution spatial transcriptomics samples. By effectively encoding molecular context to refine WSI representations, PathLUPI offers a strategy to bridge molecular insights with routine pathology workflows for wider clinical application. Obtaining accurate molecular insights for precision oncology remains costly and time consuming. Here, the authors develop PathLUPI, a deep learning method to enable effective molecular prediction in cancer using only routine whole-slide images, with evaluation across 49 molecular oncology tasks in 20 different cohorts.