QValues.jl

Installation

QValues can be installed from within Julia using

Pkg.clone("git@github.com:kdyrhage/QValues.jl.git")

Usage

QValues only exports one function:

# Calculate q-values given a set of p-values P:
qvalues(P)

# A range or vector containing lambda values to be tested can also be included:
qvalues(P, 0.01:0.01:0.99)

# The method for estimating the rate of true nulls can be changed, and for
# :bootstrap additional variables can be set:
qvalues(P, method = :bootstrap, B = 1000, fraction = 0.8)

# Other estimation methods ignore these options, so the following two are equivalent:
qvalues(P, method = :fast, B = 1, fraction = 0.5)
qvalues(P, method = :fast)

Three methods are available for estimating the null hypothesis rate: :spline, :bootstrap, and :fast. :spline and :bootstrap use the methods described in [1] and [2], respectively, while :fast uses the "bootstrap" method used in the qvalue package for R (which does not actually involve any bootstrapping). :bootstrap is much slower than the other two methods. The number of bootstraps can be adjusted with the B keyword, and the size of the fraction used for each iteration can be adjusted with fraction.

Examples usage:

using QValues
using RDatasets
using Plots

# Load a dataset that contains p-values. This set actually contains
# values in 0 < p < 0.05, so multiply by 20 to get 0 < p < 1
manhattan = dataset("gap", "mhtdata")
P = 20 * manhattan[:P]

Q = qvalues(P, 0.05:0.05:0.95, method = :bootstrap)

scatter(P, Q, xlabel = "p-value", ylabel = "q-value", marker = (2, :black))

Example usage without external packages:

using QValues

# Rough simulation of p-values for true null (p0) and true positives (p1)
P0 = rand(1000)
P1 = vcat([rand(div(1000, 2i)) / i^2 for i in 4:9]...)
P = [P0; P1]

Q = qvalues(P, method = :bootstrap)
sum(Q .<= 0.05)

References

1. John D. Storey and Robert Tibshirani, Statistical significance for genomewide studies, Proceedings of the National Academy of Sciences 100, 9440-9445 (2003)
2. John D. Storey, A direct approach to false discovery rates, Journal of the Royal Statistical Society: Series B 64, 479-498 (2002)