NBSR: a Negative Binomial Softmax Regression model for microRNA-seq data analysis

Area of Science:

• Genomics
• Bioinformatics
• Molecular Biology

Background:

• MicroRNAs (miRNAs) are key regulators of gene expression and disease.
• Statistical methods for miRNA analysis lag behind those for messenger RNA (mRNA).
• Existing mRNA-based methods applied to miRNA data may yield inaccurate results due to unique miRNA characteristics.

Purpose of the Study:

• To evaluate the suitability of mRNA sequencing methods for miRNA data.
• To propose and validate a novel statistical model for miRNA sequencing data analysis.
• To enhance the accuracy and statistical power of differential expression analysis for miRNAs.

Main Methods:

• Examination of assumptions in mRNA-based methods for miRNA analysis.
• Development and application of a negative binomial softmax regression (NBSR) model.
• Utilizing log relative abundance ratio (log-RAR) for differential expression interpretation.
• Modeling the relationship between biological coefficient of variation and relative abundance.
• Debiasing log-RAR for accurate fold change inference.

Main Results:

• mRNA-based methods can lead to high false discovery rates in miRNA analysis.
• The proposed NBSR model offers improved statistical power and narrower confidence intervals.
• NBSR effectively manages highly variable and sparsely expressed miRNAs, increasing detection sensitivity.
• Debiased log-RAR allows precise fold change estimation, even with limited differentially expressed miRNAs.
• The NBSR model demonstrated efficacy on both simulated and real-world datasets.

Conclusions:

• The NBSR model provides a more accurate and sensitive approach for miRNA sequencing data analysis.
• This method addresses limitations of existing techniques, particularly for variable and sparse data.
• NBSR facilitates robust differential expression analysis, crucial for understanding miRNA roles in disease and gene regulation.