Integrative clustering of multi-level omics data for disease subtype discovery using sequential double...

Area of Science:

• Bioinformatics
• Computational Biology
• Statistical Genetics

Background:

• Multi-level omics data integration presents statistical challenges for identifying disease subtypes.
• Accurate patient clustering is crucial for personalized medicine and tailored treatments.
• Existing methods like iCluster require improved feature selection and handling of outlier samples.

Purpose of the Study:

• To propose an improved iCluster method for multi-omics data integration and patient clustering.
• To enhance feature selection using prior knowledge of inter-omics regulatory flows.
• To develop a method capable of clustering with scattered samples and generating tight clusters.

Main Methods:

• Utilized an overlapping sparse group lasso penalty for feature selection within the iCluster factor model.
• Incorporated sample regularization to accommodate and cluster scattered samples.
• Evaluated the proposed group structured tight iCluster method using real breast cancer data and simulations.

Main Results:

• Demonstrated improved clustering accuracy compared to existing methods.
• Showcased enhanced biological interpretation of identified disease subtypes.
• Successfully generated coherent and tight clusters, effectively handling scattered samples.

Conclusions:

• The proposed group structured tight iCluster method offers a robust approach for multi-omics data integration and patient subtyping.
• This method advances the potential for personalized medicine by improving the identification of clinically relevant disease subtypes.
• The enhanced feature selection and scattered sample handling contribute to more accurate and interpretable clustering results.