Multi-modal transformer focused on pathway analysis, leveraging KEGG, PID, Reactome, and BioCarta databases. According to them: "These multi-modal data have 115 different dimensions, including nucleotide level, fragment level, and gene level. For example, Pathformer’s input for TCGA datasets includes gene-level RNA expression, fragment-level DNA methylation, and both fragment-level and gene-level DNA CNV".
The model uses criss-cross attention similar to Evoformer blocks in AlphaFold2 to fuse and select different modalities. I'm always a sucker for explainability with SHAP; they implement a "biological interpretability module" to try to decipher important pathways and key genes.