How it works

The foundation model is trained to embed histology images using genome-level signals as supervision anchors, forcing visual features to align with molecular biology rather than image patterns alone.