A Novel Algorithm for Feature Selection Using Penalized Regression with Applications to Single-Cell RNA Sequencing...

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Single-cell RNA sequencing (scRNA-seq) enables gene expression analysis at single-cell resolution.
• High dimensionality of scRNA-seq data presents significant analytical challenges.
• Machine learning offers potential for effective gene (feature) selection in scRNA-seq analysis.

Purpose of the Study:

• To rigorously compare the performance of various penalized regression methods for feature selection in scRNA-seq data.
• To identify the most effective penalized regression method for handling high-dimensional scRNA-seq datasets.
• To develop a novel algorithm for gene selection in scRNA-seq data based on penalized regression.

Main Methods:

• Comparative analysis of seven penalized regression methods: ridge, lasso, elastic net, drop lasso, group lasso, big lasso, and sparse group lasso (SGL).
• Evaluation of methods using area under the receiver operating curve (AUC) and computation time metrics.
• Development of a new gene selection algorithm employing SGL and hierarchical clustering for gene grouping.

Main Results:

• Sparse Group Lasso (SGL) demonstrated superior performance compared to the other six methods in terms of AUC and computation time.
• The proposed algorithm, utilizing SGL and hierarchical clustering, achieved consistently better AUC across analyzed scRNA-seq datasets.
• The algorithm effectively identifies differentially expressed genes without requiring domain-specific gene grouping knowledge.

Conclusions:

• SGL is a highly effective penalized regression method for feature selection in scRNA-seq data analysis.
• The proposed algorithm offers an improved and robust approach for identifying key genes in scRNA-seq studies.
• This work advances the analytical capabilities for high-dimensional single-cell gene expression data.