Independent filtering multiple testing correction

independent-filtering-multiple-testing-correction

Summary

Automatically filter low-abundance genes before multiple-testing correction in RNA-seq differential expression analysis to improve statistical power and control false discovery rate. This skill combines independent filtering (removing genes with low mean normalized counts) with FDR-based multiple-testing correction to identify significantly differentially expressed genes.

When to use

Apply this skill when analyzing RNA-seq count data from a DESeq2 workflow where you have fitted negative binomial generalized linear models and need to extract final results. Use it specifically when you want to identify genes with adjusted p-value below a predetermined significance threshold (e.g., α=0.1 for FDR) while maximizing statistical power by removing genes unlikely to show differential expression due to low abundance.

When NOT to use

• Input data has already been pre-filtered by external count threshold—independent filtering will apply an additional, data-driven filter that may not align with your prior cutoff.
• You require gene-level p-values without multiple-testing correction (e.g., for visualization or hypothesis generation only).
• Your experimental design involves a very small number of genes (e.g., targeted panel <100 genes), where independent filtering may remove too many genes and compromise power.

Inputs

• DESeqDataSet object (dds) after DESeq() analysis
• design formula (e.g., ~cell+dex)
• normalized count matrix
• sample metadata (cell type, treatment condition)

Outputs

• results table with log2 fold changes, p-values, and adjusted p-values
• filtered gene list with padj < alpha
• count of significantly differentially expressed genes

How to apply

After running DESeq(dds) to estimate dispersions and fit models, call results(dds) with your chosen alpha threshold (default alpha=0.1 implements FDR control). The results() function automatically performs independent filtering based on the mean of normalized counts for each gene—genes below a data-driven quantile cutoff are excluded before p-value adjustment. This reduces multiple-testing burden, increases sensitivity for the remaining genes, and avoids inflating type I error. Order the results table by adjusted p-value (padj) and extract genes meeting your padj < alpha threshold. The independent filtering step is performed internally by the results function and does not require manual specification of a count threshold.

Related tools

• DESeq2 (Performs negative binomial GLM fitting, independent filtering, and multiple-testing correction via results() function) — https://github.com/thelovelab/DESeq2
• ashr (Provides adaptive shrinkage methods for improved effect size estimation that can follow independent filtering and FDR correction) — https://github.com/stephens999/ashr
• DEvis (Visualizes and aggregates differential expression results including filtered gene sets) — https://github.com/price0416/DEvis

Examples

dds <- DESeq(dds); res <- results(dds, alpha=0.1); res_ordered <- res[order(res$pvalue),]; sig_genes <- res_ordered[res_ordered$padj < 0.1,]; nrow(sig_genes)

Evaluation signals

• Verify that the number of genes returned matches the reported count of genes with padj < alpha (e.g., in task_001, reproducibility of the exact count confirms correct filtering and correction).
• Check that the results table contains columns for baseMean, log2FoldChange, pvalue, padj, and padj values are strictly ≥ pvalue (indicating multiple-testing correction was applied).
• Confirm that independent filtering was active by verifying that very low-abundance genes (e.g., baseMean < 1) are absent from the results table, even if their raw p-value was significant.
• Validate that adjusted p-values are monotonically non-decreasing when results are ordered by p-value (property of independent filtering with Benjamini-Hochberg correction).
• Cross-check reproducibility: re-running results(dds, alpha=0.1) on the same dds object should yield identical gene counts and adjusted p-values.

Limitations

• Independent filtering is automatic and uses a default quantile cutoff determined from the data; no user control over the exact count threshold unless manually pre-filtering before DESeq().
• The filter is designed for data with a wide range of counts; highly uniform or heavily pre-filtered datasets may not benefit from independent filtering.
• Results are sensitive to the alpha threshold chosen; no consensus exists on whether alpha=0.05, 0.1, or other values is appropriate for a given study, and FDR control assumes independence or weak dependence among tests.

Evidence

• [other] the results function automatically performs independent filtering based on the mean of normalized counts for each gene: "the results function automatically performs independent filtering based on the mean of normalized counts for each gene"
• [other] Extract results using results(dds) with default alpha=0.1, which automatically performs independent filtering based on mean normalized counts.: "Extract results using results(dds) with default alpha=0.1, which automatically performs independent filtering based on mean normalized counts."
• [other] by removing rows in which there are very few reads, we reduce the memory size of the dds data object, and we increase the speed: "by removing rows in which there are very few reads, we reduce the memory size of the dds data object, and we increase the speed"
• [other] The package DESeq2 provides methods to test for differential expression by use of negative binomial generalized linear models: "The package DESeq2 provides methods to test for differential expression by use of negative binomial generalized linear models"