ME-Mamba: Multi-Expert Mamba with efficient knowledge capture and fusion for multimodal survival analysis.

Area of Science:

• Computational biology
• Bioinformatics
• Precision oncology

Background:

• Multimodal survival analysis integrating whole slide images (WSIs) and genomics is crucial for precision oncology.
• Current Transformer-based methods face challenges with quadratic complexity and noise susceptibility.

Purpose of the Study:

• To introduce ME-Mamba, a Multi-Expert Mamba framework designed for efficient and robust multimodal survival analysis.
• To address the computational complexity and noise issues in integrating WSIs and genomics data.

Main Methods:

• Developed a Multi-Expert Mamba framework (ME-Mamba) with linear computational complexity.
• Proposed an attention-guided scan strategy to overcome Mamba's sequential scanning bias.
• Introduced a Synergistic Expert with a parameter-free, dual-granularity fusion mechanism using Optimal Transport (OT) and Maximum Mean Discrepancy (MMD).

Main Results:

• ME-Mamba achieved linear complexity, significantly reducing computational burden.
• The attention-guided scan and Synergistic Expert effectively prioritized discriminative features and improved signal-to-noise ratio.
• Demonstrated superior prediction accuracy and computational efficiency compared to state-of-the-art methods across five TCGA datasets.

Conclusions:

• ME-Mamba offers a computationally efficient and robust solution for multimodal survival analysis in precision oncology.
• The proposed framework effectively integrates WSIs and genomics data, paving the way for improved cancer patient stratification and treatment strategies.
• The code will be made publicly available to facilitate further research.