Meta-analytic framework for sparse K-means to identify disease subtypes in multiple transcriptomic studies.

Area of Science:

• Computational biology
• Bioinformatics
• Machine learning in healthcare

Background:

• Omics data analysis is crucial for personalized medicine.
• Identifying distinct disease subtypes is key for tailored treatments.
• Unsupervised machine learning offers a pathway for disease subtyping.

Purpose of the Study:

• To develop a meta-analytic framework for identifying novel disease subtypes using multi-cohort omics data.
• To enhance the accuracy and stability of disease subtyping compared to single-study approaches.
• To enable the development of targeted therapies for specific disease subgroups.

Main Methods:

• Extension of sparse K-means clustering to a meta-analytic framework.
• Utilizing lasso regularization for identifying key gene features.
• Incorporating a pattern matching reward function for cross-study consistency.
• Validation through simulations and analysis of leukemia and breast cancer datasets.

Main Results:

• The meta-analytic framework identified novel disease subtypes with improved accuracy and stability.
• Gene features identified were unique and crucial for subtype characterization.
• Application to breast cancer data on an independent dataset showed significant survival differences between subtypes.
• The method demonstrated superior performance over single-study analyses.

Conclusions:

• The developed meta-analytic framework effectively identifies robust disease subtypes from multi-cohort omics data.
• This approach supports the development of precise diagnostic tools and targeted therapeutic strategies.
• The findings pave the way for more effective personalized treatment strategies in complex diseases.