View-Aware Collaborative Learning for Survival Prediction and Subgroup Identification.

Area of Science:

• Computational biology
• Bioinformatics
• Clinical data analysis

Background:

• High-throughput omics technologies generate diverse datasets crucial for understanding disease.
• Multi-view learning can improve survival prediction and subgroup identification in clinical analysis.
• These tasks are often studied in isolation, limiting potential synergistic benefits.

Purpose of the Study:

• To develop a novel method, View-aware Collaborative Learning (VaCoL), for jointly improving survival prediction and subgroup identification.
• To leverage multiple omics data through a unified learning framework.
• To adaptively weight different omics data views based on their importance.

Main Methods:

• Integration of survival analysis and affinity learning into a unified optimization framework.
• Utilizing the log-rank test statistic to assess the importance of different omics data views.
• Developing an adaptive view-weighting mechanism within the collaborative learning process.

Main Results:

• Empirical results on real datasets demonstrate significant performance improvements in both survival prediction and subgroup identification.
• The proposed VaCoL method effectively integrates multiple omics data for enhanced clinical analysis.
• Model analysis confirms the effectiveness of the collaborative learning and adaptive view-weighting strategies.

Conclusions:

• The VaCoL method offers a powerful approach for jointly analyzing multiple omics data to improve clinical predictions.
• Collaborative learning between survival prediction and subgroup identification enhances overall performance.
• Adaptive weighting of omics data views optimizes the integration of diverse biological information.