SMSPL: Robust Multimodal Approach to Integrative Analysis of Multiomics Data

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Advancing technologies enable diverse genome-wide data collection, facilitating biological system understanding through integrated analysis.
• Current multiomics data integration methods struggle with high noise, leading to overfitting and limited generalization.
• Sample reweighting is a common strategy to mitigate noise in multiomics integration.

Purpose of the Study:

• To propose a robust multimodal data integration method (SMSPL) for cancer subtype prediction and multiomics signature identification.
• To address the challenge of high noise in multiomics data integration.
• To improve the generalization performance of multiomics analysis.

Main Methods:

• SMSPL leverages data linkages to interactively recommend high-confidence samples.
• A novel soft weighting scheme assigns weights to training samples for each data type.
• The method iterates between recalculating weights and updating classifiers.

Main Results:

• Simulation and five real experiments demonstrate SMSPL's capability in classification.
• The method successfully identifies significant multiomics signatures even with heavy noise.
• SMSPL shows robust performance in handling noisy multiomics data.

Conclusions:

• SMSPL offers a robust approach to multimodal data integration, enhancing cancer subtype prediction and multiomics signature discovery.
• The method effectively manages noise, improving classification and generalization.
• SMSPL represents a step forward in multiomics data integration for comprehensive biological understanding.