qvalue
Dependencies & Environment
Package-intrinsic requirements from the Bioconductor landing page — reproduce in any R environment.
- Version: 2.44.0 · Bioconductor: 3.23 · R: ≥ 4.6
- Imports: ggplot2, reshape2
- Install:
BiocManager::install("qvalue")
When to Use
- False Discovery Rate Estimation: When analyzing thousands of features (e.g., differential gene expression) to estimate q-values and control the false discovery rate using
qvalue. - Estimating True Nulls: When you need to estimate the overall proportion of true null hypotheses ($\pi_0$) from a distribution of p-values using
pi0est. - Local FDR Calculation: When you want to calculate the empirical Bayesian posterior probability that a null hypothesis is true, conditional on the observed p-value, using
lfdr. - Empirical P-value Calculation: When you have observed test-statistics and simulated/resampled null statistics and need to compute p-values using
empPvals.
When NOT to Use
- Dependent Test Statistics: When tests are strongly dependent or have uncorrected batch effects (indicated by a U-shaped p-value histogram), use the
svapackage first to correct the data. - Small-scale hypothesis testing: For experiments with very few tests, use standard Benjamini-Hochberg adjustment because the estimation of $\pi_0$ is highly unstable with small sample sizes.
Data Requirements
- A numeric vector of p-values strictly bounded between 0 and 1, OR a vector of observed statistics and a matrix of empirical null statistics.
- P-values must follow a Uniform(0,1) distribution under the null hypothesis (the right tail of the p-value histogram should be fairly flat).
Key Parameters
- p (required): A vector of p-values.
- fdr.level (NULL): The level at which to control the false discovery rate; returns a logical vector indicating significant tests.
- pfdr (FALSE): Indicator of whether to use the positive false discovery rate (pFDR) to make the estimate more robust for small p-values.
- lambda (seq(0, 0.95, 0.05)): Values of the tuning parameter considered in estimating $\pi_0$.
- pi0.method ("smoother"): Method for automatically handling the tuning parameter in $\pi_0$ estimation ("smoother" or "bootstrap").
- trunc (TRUE): If TRUE, local FDR estimates > 1 are set to 1 in
lfdr. - stat (NULL): Vector of observed test-statistics calculated on the original data for
empPvals. - stat0 (NULL): Matrix of empirical null statistics for
empPvals.
Best Practices
- Check P-value Histogram: Always view a histogram of the p-values (
hist(p)) before runningqvalueto confirm the right tail is flat. - Evaluate Pi0 Reliability: Use
plot(qobj)to view the estimated $\pi_0$ versus the tuning parameter $\lambda$ to gauge the reliability of the $\pi_0$ estimate. - Summarize Results: Use
summary(qobj)to view the $\pi_0$ estimate and the cumulative number of significant calls at various p-value, q-value, and local FDR cutoffs.
Common Pitfalls
- U-shaped p-value histograms: Indicates dependence among variables or a one-sided test on data with two-sided signal. Fix: Compute p-values using a different model or use the
svapackage to correct for dependence. - Assuming q-values are adjusted p-values: Users often worry when q-values are smaller than original p-values. Fix: Understand that q-values are population quantities bounded by $\pi_0$, not Bonferroni adjusted p-values.
- Incorrect empirical p-values: Passing statistics where larger values do not indicate more evidence against the null. Fix: Ensure statistics passed to
empPvalsare constructed such that larger values are "more extreme" (e.g., absolute values of t-statistics).
Alternatives
- sva: For correcting dependent variables and batch effects before computing p-values.
- fdrtool: Estimates local FDR and tail area FDR directly from z-scores or p-values.
- IHW: Incorporates covariates (like base mean expression) to increase statistical power.
- ashr: Uses adaptive shrinkage to model the distribution of effects, providing robust local FDR estimates.
Citations
- Storey JD. (2002). A direct approach to false discovery rates. Journal of the Royal Statistical Society, Series B, 64:479–498.
- Storey JD. (2003). The positive false discovery rate: A Bayesian interpretation and the q-value. Annals of Statistics, 31:2013–2035.
- Storey JD, Tibshirani R. (2003). Statistical significance for genome-wide experiments. Proceedings of the National Academy of Sciences, 100:9440–9445.
References
- Homepage: https://bioconductor.org/packages/qvalue
- Vignette: https://bioconductor.org/packages/release/bioc/vignettes/qvalue/inst/doc/qvalue.pdf
Run this on BioMate
This skill is the knowledge layer — when, why, and how to use qvalue. To run this analysis on your own data with managed compute, automated QC, and reproducible outputs, use BioMate — free to start.
