UDEC-MO: an uncertainty-guided deep embedded clustering framework for bulk and single-cell multi-omics data

Area of Science:

• Computational biology
• Bioinformatics
• Data science

Background:

• Multi-omics technologies offer comprehensive molecular insights into disease heterogeneity and cellular diversity.
• Robust clustering of multi-omics data is hindered by high dimensionality, noise, and varying data quality across features, modalities, and samples.

Purpose of the Study:

• To develop a novel framework, UDEC-MO (Uncertainty-guided Deep Embedded Clustering for Multi-Omics), for robust multi-omics clustering.
• To address the challenge of multilevel data quality variation in existing clustering methods.

Main Methods:

• UDEC-MO estimates feature-wise heteroscedastic uncertainty via uncertainty-aware reconstruction.
• It calculates modality-level and instance-level uncertainty scores.
• Instance-level uncertainty is converted to reliability weights to modulate the Kullback-Leibler-divergence loss, prioritizing reliable samples.

Main Results:

• UDEC-MO demonstrated competitive or superior clustering performance on both bulk and single-cell multi-omics datasets.
• The framework effectively reduces the influence of uncertain instances during cluster refinement.
• Uncertainty-derived reliability indicators were generated for features, modalities, and instances.

Conclusions:

• UDEC-MO provides a robust approach for multi-omics clustering by incorporating uncertainty quantification.
• The method enhances the characterization of disease heterogeneity and cellular diversity.
• The derived reliability indicators offer valuable auxiliary information for data interpretation.