TMODINET: A trustworthy multi-omics dynamic learning integration network for cancer diagnostic.

Area of Science:

• Biomedical informatics
• Computational biology
• Precision medicine

Background:

• Multi-omics data offers rich biomedical insights but integration for cancer diagnosis often lacks interpretability and reliability.
• Existing methods prioritize classification accuracy over understanding internal mechanisms, crucial for clinical applications.

Purpose of the Study:

• To develop a trustworthy multi-omics dynamic learning framework (TMODINET) for enhanced cancer diagnosis.
• To improve patient-centered diagnosis by integrating multi-omics data with a focus on reliability and interpretability.

Main Methods:

• Employed adaptive dynamic learning with self-attentional mechanisms for feature and modality processing.
• Introduced a graph dynamic learning method with adaptive graph structure adjustment for graph convolutional networks (GCN).
• Utilized uncertainty quantification via Dirichlet distribution and Dempster-Shafer theory for decision-level data integration.

Main Results:

• The TMODINET framework demonstrated superior performance and trustworthiness compared to state-of-the-art methods.
• Extensive experiments on four real-world multimodal medical datasets validated the model's effectiveness.
• The proposed approach provides patient-centered diagnoses with enhanced reliability.

Conclusions:

• The TMODINET framework offers a reliable and interpretable approach to multi-omics data integration for cancer diagnosis.
• The model shows significant potential for clinical applications in precision medicine and life sciences.
• Integrating uncertainty quantification enhances the trustworthiness of multi-omics-based diagnostic models.