An uncertainty-aware dynamic decision framework for progressive multi-omics integration in classification tasks.

Area of Science:

• Computational biology
• Bioinformatics
• Precision medicine

Background:

• High-throughput multi-omics profiling is crucial for early disease diagnosis.
• Challenges include missing coordinated molecular interactions and high costs of comprehensive profiling.

Purpose of the Study:

• To develop an uncertainty-aware, multi-view dynamic decision framework.
• To enhance classification accuracy and reduce diagnostic testing costs.

Main Methods:

• Incorporating subjective logic and Dirichlet distributions for uncertainty estimation at the single-omics level.
• Fusing complementary omics modalities using Dempster-Shafer theory for multi-omics analysis.
• Implementing a dynamic decision mechanism for incremental omics view integration based on real-time uncertainty.

Main Results:

• Accurate classification achieved using a single omics modality in over 50% of cases across benchmark datasets (ROSMAP, LGG, BRCA, KIPAN).
• Maintained diagnostic performance comparable to full-omics models.
• Effectively reduced redundant testing and preserved biological insights.

Conclusions:

• The framework enables an 'on-demand testing' paradigm in precision medicine.
• Facilitates intelligent resource allocation and reduces healthcare costs, particularly in resource-limited settings.