RankByGene: Gene-Guided Histopathology Representation Learning Through Cross-Modal Ranking Consistency
Researchers have developed a new framework called RankByGene to improve the alignment between spatial transcriptomics (ST) data and histology images. This method uses a novel ranking-based alignment loss to preserve relative similarities across modalities, addressing challenges like spatial distortions and modality-specific variations. The framework also incorporates self-supervised knowledge distillation with a teacher-student network to enhance alignment stability, particularly with noisy gene expression data. Experiments across seven datasets show RankByGene outperforms existing methods in alignment and predictive tasks, including gene expression prediction, slide-level classification, and survival analysis. AI
IMPACT Enhances biological data analysis by improving cross-modal alignment for better gene expression prediction and survival analysis.