Polyfit

Polyfit corrects non-uniform p-value distributions in differential gene expression analyses to provide accurate false discovery rate (FDR) estimation for RNA-sequencing two-class experiments.

Key Features:

• Integration with DESeq and edgeR: Acts as an add-on to process and adjust differential expression results from DESeq and edgeR.
• Non-uniform p-value correction: Addresses non-uniform p-value distributions that can bias FDR estimates and reduce power to detect true differential expression.
• Uniform null p-value distribution: Ensures the distribution of p-values under the null hypothesis is uniform on the interval [0, 1].
• Adapted Storey-Tibshirani method: Implements an adapted Storey-Tibshirani approach to calculate q-values for FDR control.
• Two-class dataset focus: Targets two-class differential expression experiments.
• Validation on synthetic and real data: Validated using both synthetic datasets and real biological RNA-seq data.
• Replicate-aware performance: Demonstrates improved performance with six or more biological replicates per condition.
• Improves DESeq performance: Improves FDR estimation for DESeq and makes its performance comparable to edgeR and DESeq2.
• Accurate low FDRs across p-value range: Achieves low false discovery rates that are accurately estimated across the full range of p-values.

Scientific Applications:

• RNA-seq differential expression: Provides corrected p-values and q-values for two-class RNA-seq differential gene expression analyses.
• FDR control in replicate-rich studies: Improves FDR estimation and detection power in studies with six or more biological replicates per condition.
• Post-processing of DE results: Can be applied as a post-processing step to DESeq and edgeR outputs to adjust p-value distributions and q-values.
• Comparative method benchmarking: Enables more reliable comparison of DESeq results with edgeR and DESeq2 by correcting p-value distributions.

Methodology:

Adapts the Storey-Tibshirani q-value calculation and enforces uniform null p-value distributions for two-class datasets; validated on synthetic and real biological RNA-seq data.