CAEN

CAEN selects feature genes for classifying bulk and single-cell RNA-seq data by using rank-based category encoding and correlation analysis to identify differentially expressed (DE) genes that distinguish classes.

Key Features:

• Category Encoding via Rank: Encodes categories using the rank of sequence samples for each gene across different classes to represent class-specific expression patterns.
• Correlation Coefficients Analysis: Computes correlation coefficients between genes and classes based on sample ranks and a defined category rank to quantify association strength.
• Feature Gene Identification: Selects genes with the highest correlation coefficients as feature genes for downstream classification and dimensionality reduction.
• Sure Screening Method: Applies a sure screening approach to ensure rank consistency properties among selected features.
• Nonparametric Design (vs BSS/WSS): Operates without assuming normal data distributions, diverging from BSS/WSS (BW) approaches that rely on parametric assumptions.
• Performance Evaluation: Demonstrates classifier performance through simulation studies and analyses of real datasets, showing comparable or superior results to competing methods in most settings.

Scientific Applications:

• RNA-seq classification: Improves classification of bulk and single-cell RNA-seq samples by identifying discriminative feature genes.
• Disease diagnosis: Enables selection of gene sets that can distinguish disease states from controls in expression-based diagnostic studies.
• Cellular heterogeneity analysis: Facilitates discovery of genes that separate cellular subpopulations in single-cell RNA-seq datasets.
• Molecular mechanism exploration: Supports investigation of genes underlying different biological conditions by highlighting DE feature genes relevant to class differences.

Methodology:

CAEN performs rank-based category encoding, computes correlation coefficients between genes and classes using sample ranks and a category rank, selects genes with highest correlations, and applies a sure screening step to ensure rank consistency while avoiding assumptions of normality (contrasting with BSS/WSS (BW)).