The paper introduces a Multimodal Prototyping framework (MMP) for cancer survival prediction, which combines gigapixel whole-slide images (WSIs) and transcriptomic profiles. Traditional methods tokenize WSIs into thousands of patches and transcriptomics into gene groups, leading to high computational costs and interpretability challenges. MMP proposes using *morphological* prototypes to compress WSIs by over 300× and *biological pathway* prototypes to encode transcriptomic profiles, both in an unsupervised manner. These prototypes are then processed by a fusion network, either a Transformer or optimal transport (OT) cross-alignment, reducing the number of tokens and improving efficiency. Extensive evaluation on six cancer types shows that MMP outperforms state-of-the-art methods with less computation and enhanced interpretability. The framework is available at <https://github.com/mahmoodlab/MMP>.The paper introduces a Multimodal Prototyping framework (MMP) for cancer survival prediction, which combines gigapixel whole-slide images (WSIs) and transcriptomic profiles. Traditional methods tokenize WSIs into thousands of patches and transcriptomics into gene groups, leading to high computational costs and interpretability challenges. MMP proposes using *morphological* prototypes to compress WSIs by over 300× and *biological pathway* prototypes to encode transcriptomic profiles, both in an unsupervised manner. These prototypes are then processed by a fusion network, either a Transformer or optimal transport (OT) cross-alignment, reducing the number of tokens and improving efficiency. Extensive evaluation on six cancer types shows that MMP outperforms state-of-the-art methods with less computation and enhanced interpretability. The framework is available at <https://github.com/mahmoodlab/MMP>.