Bayesian negative binomial regression for differential expression with confounding factors.

Area of Science:

• Genomics
• Bioinformatics
• Statistical Genetics

Background:

• High-throughput sequencing advances understanding of genome-wide molecular profiles.
• Existing methods often analyze single factors and require normalization, limiting applicability.
• Complex multi-factor experiments necessitate advanced statistical tools for genotype-phenotype relationship analysis.

Purpose of the Study:

• Develop a novel statistical method for RNA sequencing (RNA-seq) count data analysis.
• Address limitations of existing methods in handling complex experimental designs and confounding factors.
• Provide a robust tool for deciphering genotype-phenotype relationships in multi-factor studies.

Main Methods:

• Developed a Bayesian negative binomial regression (BNB-R) method.
• Utilized natural model parameterization to eliminate the need for normalization.
• Employed data augmentation techniques for efficient Bayesian inference of model parameters.
• Designed to handle complex experimental designs with multi-variate dependencies.

Main Results:

• BNB-R demonstrates superior performance in analyzing RNA-seq data.
• The method effectively handles complex experimental designs and multi-factor interactions.
• Evaluations on synthetic and real-world data show improved accuracy using ROC and precision-recall curves.

Conclusions:

• BNB-R offers a powerful and adaptable solution for RNA-seq data analysis.
• The method simplifies analysis by removing the need for normalization.
• BNB-R facilitates accurate genotype-phenotype relationship studies in complex biological systems.