Categorical Data Analysis for High-Dimensional Sparse Gene Expression Data

Area of Science:

• Biostatistics
• Bioinformatics
• Genomics

Background:

• Analyzing categorical data with high-dimensional sparse covariates, common in omics data, presents significant challenges.
• Existing statistical methods may not adequately address the complexities of variable and model selection in such datasets.

Purpose of the Study:

• To introduce a comprehensive statistical procedure for categorical data analysis in the context of high-dimensional omics data.
• To enable variable screening, model selection, response category ordering, and variable selection for complex biological datasets.

Main Methods:

• A statistical procedure based on multinomial logistic regression analysis was developed.
• The procedure incorporates variable screening, model selection, order selection for response categories, and variable selection.
• The method was applied to high-dimensional gene expression data from 801 patients across five cancerous tumor types.

Main Results:

• A finalized model with 74 genes demonstrated extremely low cross-entropy loss and zero predictive error rate via five-fold cross-validation.
• Two additional models, comprising 31 and 4 genes, were identified as potential prognostic multi-gene signatures.

Conclusions:

• The proposed statistical procedure is effective for categorical data analysis with high-dimensional sparse covariates, particularly in omics research.
• The identified gene signatures offer potential for improved cancer tumor classification and prognosis.