Block HSIC Lasso: model-free biomarker detection for ultra-high dimensional data

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Identifying non-linear relationships between biomolecules and biological outcomes is computationally intensive and statistically complex.
• Current feature selection methods often suffer from drawbacks like lack of parsimony, non-convexity, and high computational cost.

Purpose of the Study:

• To introduce block HSIC Lasso, a novel non-linear feature selection method designed to overcome the limitations of existing approaches.
• To evaluate the performance of block HSIC Lasso against state-of-the-art feature selection techniques.

Main Methods:

• Block HSIC Lasso, a new non-linear feature selection algorithm.
• Comparative analysis using synthetic and real-world genomic datasets (gene-expression microarrays, single-cell RNA sequencing, genome-wide association studies).

Main Results:

• Block HSIC Lasso demonstrates superior performance in retaining biological information compared to other methods across various genomic data types.
• Application to single-cell RNA sequencing data from mouse hippocampus identified genes crucial for neuronal development and function.

Conclusions:

• Block HSIC Lasso offers an efficient and effective solution for non-linear feature selection in biological data analysis.
• The method provides deeper biological insights, particularly in complex datasets like single-cell RNA sequencing.