Missing-Modality-Aware Graph Neural Network for Cancer Classification

In three linesMAGNET, a graph neural network, handles incomplete multimodal biological data for cancer classification. The model uses a dynamic multi-head attention mechanism to fuse modality embeddings with missing patterns, achieving linear complexity. Tested on three public multiomics datasets, MAGNET outperforms existing fusion methods.

Summary generated by Claude — human-verified