Gene Expression Prediction from Histopathology Images

This field predicts gene activity from stained tissue images, either across whole slides or at individual spots and cells, to reveal molecular patterns while reducing reliance on costly sequencing. Common approaches combine pretrained pathology image encoders with spatial context models such as transformers and graphs, or use cell-type structure, gene programs, retrieval, and generative models to capture biological relationships. Results increasingly support spatial expression mapping and downstream tasks such as pathway analysis and treatment-response prediction, but performance can vary across tissues, patients, and sequencing platforms. A central challenge is establishing reliable biological value through standardized, patient-level evaluation that tests generalization and spatial fidelity, not just average prediction accuracy.