sSeq

sSeq implements shrinkage-based dispersion estimation to identify differential gene expression from RNA-seq count data modeled by the Negative Binomial distribution.

Key Features:

• Negative Binomial modeling: Models RNA-seq read counts using the Negative Binomial (NB) distribution to separate systematic expression changes from noise.
• Small-sample dispersion estimation: Addresses unreliable per-gene dispersion estimates in small sample sizes by using regularization.
• Initial estimation: Obtains initial per-gene dispersion estimates using the method of moments.
• Shrinkage regularization: Shrinks initial dispersion estimates toward a common value that minimizes the average squared difference between the initial and shrunk estimates.
• No additional modeling assumptions: Implements the shrinkage approach without requiring extra modeling assumptions.
• Compatibility with exact tests: Produces dispersion estimates that are compatible with exact tests of differential expression.
• Computational simplicity and efficiency: Uses a straightforward computation for dispersion estimation and demonstrates computational efficiency in small-sample experiments.
• Benchmarking: Evaluated on simulated and experimental datasets and shown to improve sensitivity, specificity, and computational efficiency relative to edgeR, DESeq, baySeq, BBSeq, and SAMseq in small-sample settings.

Scientific Applications:

• Differential expression between two conditions: Identification of genes with differential expression between two experimental conditions in RNA-seq studies.
• Small-sample RNA-seq studies: Analysis of experiments with limited biological replicates where per-gene dispersion estimates are unstable.
• Exact-test-based inference: Providing dispersion estimates suitable for use with exact tests of differential expression to support rigorous statistical inference.

Methodology:

Models counts with the Negative Binomial distribution, obtains initial per-gene dispersions via the method of moments, and applies shrinkage of those estimates toward a common value chosen to minimize the average squared difference between initial and shrunk estimates; results are compatible with exact tests of differential expression.