Information-incorporated sparse convex clustering for disease subtyping.

Area of Science:

• Computational biology
• Bioinformatics
• Genomics

Background:

• Human diseases exhibit heterogeneity, complicating accurate characterization and treatment.
• High-throughput multi-omics data offer potential for understanding disease mechanisms and heterogeneity.
• Existing literature contains valuable information for disease subtyping, but current clustering methods like Sparse Convex Clustering (SCC) do not fully utilize this prior knowledge.

Purpose of the Study:

• To develop an information-incorporated clustering procedure for improved disease subtyping in precision medicine.
• To leverage existing literature information and multi-omics data for enhanced disease characterization.
• To improve biomarker identification for clinical applications.

Main Methods:

• Developed an information-incorporated Sparse Convex Clustering (SCC) method.
• Utilized text mining to incorporate prior information from published studies via a group lasso penalty.
• Integrated heterogeneous data, including multi-omics data.
• Conducted simulation studies to evaluate performance under varying prior information accuracy.
• Applied the method to breast and lung cancer omics data.

Main Results:

• The proposed method significantly outperforms existing clustering techniques (SCC, K-means, Sparse K-means, iCluster+, Bayesian Consensus Clustering).
• Achieved more accurate disease subtyping compared to traditional methods.
• Successfully identified important biomarkers in real-world cancer omics data.
• Demonstrated effective utilization of prior information and multi-omics data.

Conclusions:

• The information-incorporated clustering procedure enables coherent pattern discovery and feature selection.
• This method enhances disease subtyping and biomarker identification, advancing precision medicine.
• The approach effectively integrates diverse data sources for a more comprehensive understanding of disease heterogeneity.