ShrinkBayes: a versatile R-package for analysis of count-based sequencing data in complex study designs

Area of Science:

• Genomics and Bioinformatics
• Statistical Genetics
• Computational Biology

Background:

• Observational clinical studies increasingly utilize sequencing data, presenting analytical challenges like excess zeros and complex designs.
• Small sample sizes in Bayesian inference can lead to overly liberal results or lack of power without careful prior specification or information borrowing.

Purpose of the Study:

• To introduce ShrinkBayes, a software package designed to address the analytical challenges of complex sequencing data in clinical studies.
• To demonstrate the utility of ShrinkBayes for multi-parameter inference and small sample size settings.

Main Methods:

• Application of ShrinkBayes to microRNA sequencing data from a clinical cancer study.
• Data-based simulation to evaluate multi-parameter inference performance.
• Utilizing Gaussian mixture priors with a point mass for small sample size analysis.

Main Results:

• ShrinkBayes successfully handles challenges in sequencing data analysis, including excess zeros and random effects.
• The software demonstrates strong performance in multi-parameter inference for groups.
• In small sample settings, ShrinkBayes shows high statistical power and improved false discovery rate (FDR) estimation.

Conclusions:

• ShrinkBayes is a versatile software package for analyzing count-based sequencing data.
• It is particularly beneficial for studies with complex designs and small sample sizes.
• The software enhances the reliability and power of statistical inference in genomic studies.