Large scale maximum average power multiple inference on time-course count data with application to RNA-seq analysis

Area of Science:

• Genomics
• Bioinformatics
• Statistical Biology

Background:

• Longitudinal RNA sequencing (RNA-seq) data offers insights into dynamic gene patterns.
• Existing methods for identifying differentially expressed (DE) genes in time-course data have limitations in power, theoretical grounding, and hypothesis testing flexibility.
• Current approaches may also struggle with controlling the false discovery rate.

Purpose of the Study:

• To develop a novel statistical model and testing procedure for analyzing time-course RNA-seq count data.
• To address limitations of existing methods, including improved power, theoretical justification, and the ability to test composite hypotheses.
• To ensure robust control of the false discovery rate.

Main Methods:

• A negative binomial model is proposed, conditional on a latent Gaussian mixture with evolving means.
• A general testing framework is introduced, achieving maximum average power optimality.
• The model's identifiability is established, and efficient algorithms are implemented for practical application.

Main Results:

• The new method demonstrates superior performance in simulation studies compared to existing approaches.
• The developed test effectively identifies traditional DE genes and supports a wider range of composite hypotheses.
• Application to marine diatom data revealed significant biological insights into light environment effects.

Conclusions:

• The proposed model and testing procedure offer a powerful and flexible new tool for analyzing time-course RNA-seq data.
• This method enhances the ability to uncover dynamic gene expression patterns and test complex biological questions.
• The approach has demonstrated utility in real-world biological research, providing valuable insights into physiological responses.