Category encoding method to select feature genes for the classification of bulk and single-cell RNA-seq data

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Bulk and single-cell RNA sequencing (scRNA-seq) are crucial for biological and medical research, particularly for identifying differentially expressed (DE) genes.
• Accurate classification of RNA-seq data relies on identifying informative feature genes, but many genes are not DE and can hinder performance.
• Existing methods, like BSS/WSS, often assume normal distribution, which may not be suitable for diverse RNA-seq datasets.

Purpose of the Study:

• To propose a novel Category Encoding (CAEN) method for selecting feature genes in bulk and scRNA-seq data classification.
• To enhance classification performance and computational efficiency by removing irrelevant genes.
• To provide a robust and widely usable R package for the CAEN method.

Main Methods:

• The CAEN method encodes categories using the rank of sequence samples for each gene within each class.
• It calculates correlation coefficients between genes, classes, sample ranks, and a new category rank.
• Feature genes are selected based on the highest gene correlation coefficients, with a sure screening method ensuring rank consistency.

Main Results:

• Simulation studies indicate that classifiers using the CAEN method outperform or match existing methods in most scenarios.
• Analysis of real-world datasets confirms the superior performance of CAEN compared to current competing methods.
• The CAEN method demonstrates effectiveness in identifying the most informative genes for accurate data classification.

Conclusions:

• The proposed CAEN method offers a superior approach for feature gene selection in RNA-seq data classification.
• CAEN addresses limitations of existing methods by not assuming data normality and improving efficiency.
• The availability of the R package 'CAEN' facilitates the adoption and application of this advanced method in research.