Incorporating biological information in sparse principal component analysis with application to genomic data

Area of Science:

• Bioinformatics
• Computational Biology
• Statistical Genetics

Background:

• Sparse Principal Component Analysis (PCA) is vital for high-dimensional data analysis.
• Biological networks, represented as graphs, offer valuable prior information.
• Integrating biological network data into PCA remains underexplored.

Purpose of the Study:

• Introduce novel sparse PCA methods: Fused and Grouped sparse PCA.
• Enable the incorporation of prior biological network information into variable selection.
• Enhance feature selection and prediction performance in high-dimensional data analysis.

Main Methods:

• Developed Fused sparse PCA and Grouped sparse PCA algorithms.
• Incorporated graph-based biological information into the sparse PCA framework.
• Evaluated methods through simulation studies and application to a glioblastoma dataset.

Main Results:

• Proposed methods demonstrate superior sensitivity and specificity compared to existing sparse PCA techniques.
• Methods show robustness to inaccuracies in specified graph structures.
• Application to glioblastoma data identified relevant biological pathways.

Conclusions:

• Fused and Grouped sparse PCA effectively integrate biological information for variable selection.
• Improved feature selection and interpretable principal component loadings are achieved.
• These methods offer potential insights into the molecular basis of complex diseases.