Joint between-sample normalization and differential expression detection through ℓ₀-regularized regression.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• RNA-seq data analysis requires normalization for accurate differential expression (DE) analysis.
• Existing methods often normalize before DE analysis, limiting flexibility.
• Jiang and Zhan's method jointly models normalization and DE analysis with an ℓ₀ penalty for sparse gene selection.

Purpose of the Study:

• Generalize Jiang and Zhan's method to accommodate continuous experimental variables.
• Enable detection of genes associated with single or multiple covariates.
• Develop an efficient algorithm for fitting the complex, high-dimensional model.

Main Methods:

• Extension of a joint statistical model for simultaneous normalization and DE analysis.
• Incorporation of continuous covariates into the model framework.
• Development of an efficient algorithm to handle high-dimensional, non-convex optimization problems.

Main Results:

• The generalized method successfully detects genes associated with continuous experimental variables.
• Simulations show improved detection accuracy compared to existing methods, particularly for asymmetric DE patterns and larger sample sizes.
• The method was applied to identify genes linked to pre-operative prostate-specific antigen (PSA) levels in prostate cancer patients.

Conclusions:

• The proposed method offers a powerful extension for RNA-seq DE analysis with continuous variables.
• It demonstrates superior performance in identifying DEGs under challenging data conditions.
• The application to prostate cancer data highlights its utility in real-world biological research.