Identification of relevant subtypes via preweighted sparse clustering.

Area of Science:

• Biostatistics
• Bioinformatics
• Computational Biology

Background:

• Cluster analysis is crucial for identifying homogeneous subgroups in data.
• Biomedical research often requires identifying subgroups linked to specific outcomes.
• Conventional clustering methods may fail to detect these outcome-associated subgroups, especially with high-variance features.

Purpose of the Study:

• To introduce a modified sparse clustering method for identifying biologically interesting subgroups.
• To enable the detection of secondary clusters associated with a particular outcome of interest.
• To overcome limitations of conventional clustering in high-dimensional biomedical data.

Main Methods:

• A modified sparse clustering algorithm was developed.
• The method was evaluated using simulation scenarios.
• The approach was applied to a temporomandibular disorders cohort study and a leukemia microarray dataset.

Main Results:

• The modified sparse clustering method successfully identified relevant subgroups in simulations.
• The method demonstrated efficacy in real-world biomedical datasets.
• It effectively identified clusters associated with outcomes of interest, outperforming conventional approaches.

Conclusions:

• Modified sparse clustering is a powerful tool for discovering outcome-associated subgroups in biomedical data.
• This method enhances the identification of biologically meaningful patterns.
• It offers a valuable alternative for analyzing complex, high-dimensional datasets in health research.