cma.fdr: Gene-level Empirical Bayes (EB) false discovery rate (FDR) analysis for somatic mutations in cancer

Description

Empirical Bayes estimates of the False Discovery Rate (FDR) and passenger probabilities in the analysis of somatic mutations in cancer.

Usage

cma.fdr(cma.alter, cma.cov, cma.samp, scores = c("CaMP", "logLRT"), passenger.rates = t(data.frame(.55*rep(1.0e-6,25))), allgenes=TRUE, estimate.p0=FALSE, p0.step=1, p0=1, eliminate.noval=FALSE, filter.threshold=0,  filter.above=0, filter.below=0, filter.mutations=0,  aa=1e-10,  bb=1e-10, priorH0=1-500/13020,  prior.a0=100, prior.a1=5, prior.fold=10, M=2, DiscOnly=FALSE, PrevSamp="Sjoeblom06", KnownCANGenes=NULL, showFigure=FALSE, cutoffFdr=0.1)

Arguments

cma.alter

Data frame with somatic mutation information, broken down by gene, sample, screen, and mutation type. See GeneAlterBreast for an example.

cma.cov

Data frame with the total number of nucleotides "at risk" ("coverage"), broken down by gene, screen, and mutation type. See GeneCovBreast for an example.

cma.samp

Data frame with the number of samples analyzed, broken down by gene and screen. See GeneSampBreast for an example.

scores

Vector with the scores which are to be computed. It can include: CaMP (Cancer Mutation Prevalence score), logLRT (log Likelihood Ratio Test score), neglogPg, logLRT, logitBinomialPosteriorDriver, PoissonlogBF, PoissonPosterior, Poissonlmlik0, Poissonlmlik1

passenger.rates

Data frame of passenger mutation rates per nucleotide, by type, or "context". If two rows are present, the first refers to the Discovery screen and the second to the Prevalence screen.

allgenes

If TRUE, genes where no mutations were found are considered in the analysis.

estimate.p0

If TRUE, estimates the percent of genes with only passenger mutations. Requires allgenes=TRUE

p0.step

Size of bins of histograms in the distribution of scores, to use in estimating p0 if estimate.p0 = TRUE. All scores are in the log 10 scale.

Proportion of genes with only passenger mutations. Only used if estimate.p0=FALSE

eliminate.noval

If TRUE, the genes which are not validated are eliminated from the analysis. Validated genes are those where at least one mutation was found in both the Discovery and Prevalence (or Validation) screens.

filter.threshold

This and the following three input control filtering of genes, allowing to exclude genes from analysis, by size and number of mutations. Different criteria can be set above and below this threshold. The threshold is a gene size in base pairs.

filter.above

Minimum number of mutations per Mb, applied to genes of size greater than threshold.size.

filter.below

Minimum number of mutations per Mb, applied to genes of size lower than threshold.size.

filter.mutations

Only consider genes whose total number of mutations is greater than or equal to filter.mutations.

Hyperparameter of beta prior used in compute.binomial.posterior.

priorH0

Prior probability of the null hypothesis, used to convert the BF in compute.poisson.BF to a posterior probability

prior.a0

Shape hyperparameter of gamma prior on passenger rates used in compute.poisson.BF

prior.a1

Shape hyperparameter of gamma prior on non-passenger rates used in compute.poisson.BF

prior.fold

Hyperparameter of gamma prior on non-passenger rates used compute.poisson.BF. The mean of the gamma is set so that the ratio of the mean to the passenger rate is the specified prior.fold in each type.

The number of null datasets generated to get the false discovery rates. Numbers on the order of 100 are recommended, but this will cause the function to run very slowly.

DiscOnly

If TRUE, only considers data from Discovery screen.

PrevSamp

If "Sjoeblom06", then the experimental design from Sjoeblom et al. or Wood et al. is used, namely, genes "pass" from the Discovery into the Prevalence (or Validation) screens if they are mutated at least once in the Discovery samples. If "Parsons11", the experimental design from Parsons et al. 2011 is approximated, namely, in the null datasets, a gene passes into the Prevalence screen if it is mutated at least once, and is found on a specified list of known cancer candidate (CAN) genes, or if it is mutated at least twice.

KnownCANGenes

Vector of known CAN genes, to be used if PrevSamp is not set to "Sjoeblom07".

showFigure

If TRUE, displays a figure for each score in scores, showing the right tail of the density of scores under the null, the right tail of the density of real scores as a rug (1-d) plot and the number of real genes and average number of null genes to the right of the cutoff chosen based on cutoffFdr.

cutoffFdr

If showFigure is set to TRUE, it gives the value at which we are interested in controlling the false discovery rate (Fdr). The corresponding score threshold is plotted on the figure, with the number of real genes greater than it and the average number of null genes greater than it specified. The estimated Fdr at that threshold is the ratio of the average number of null genes and the number of real genes, multiplied by p0, which is often taken to be 1.

Value

score: The score requested (e.g. the LRT).
F: Number of genes experimentally observed to give a larger score than the gene in question.
F0: Number of genes giving a larger score than the gene in question in datasets simulated from passenger mutation rates.
Fdr: The Empirical Bayes False Discovery Rate, as defined in Efron and Tibshirani 2002.
fdr: The Empirical Bayes Local False Discovery Rate, as defined in Efron and Tibshirani 2002.
p0: Scalar, Proportion of genes with only passenger mutations. Estimated or passed on from input (depending on whether estimate.p0 is TRUE

References

Efron B, Tibshirani R. Empirical Bayes methods and false discovery rates for microarrays. Genetic Epidemiology. DOI: 10.1002/gepi.1124 Parmigiani G, Lin J, Boca S, Sjoeblom T, Kinzler KW, Velculescu VE, Vogelstein B. Statistical methods for the analysis of cancer genome sequencing data, 2007. http://www.bepress.com/jhubiostat/paper126/

Sjoeblom T, Jones S, Wood LD, Parsons DW, Lin J, Barber T, Mandelker D, Leary R, Ptak J, Silliman N, et al. The consensus coding sequences of breast and colorectal cancers. Science. DOI: 10.1126/science.1133427

Wood LD, Parsons DW, Jones S, Lin J, Sjoeblom, Leary RJ, Shen D, Boca SM, Barber T, Ptak J, et al. The Genomic Landscapes of Human Breast and Colorectal Cancer. Science. DOI: 10.1126/science.1145720

Parsons DW, Jones S, Zhang X, Lin JCH, Leary RJ, Angenendt P, Mankoo P, Carter H, Siu I, et al. An Integrated Genomic Analysis of Human Glioblastoma Multiforme. Science. DOI: 10.1126/science.1164382 Jones S, Zhang X, Parsons DW, Lin JC, Leary RJ, Angenendt P, Mankoo P, Carter H, Kamiyama H, Jimeno A, et al. Core signaling pathways in human pancreatic cancers revealed by global genomic analyses. Science. DOI: 10.1126/science.1164368 Parsons DW, Li M, Zhang X, Jones S, Leary RJ, Lin J, Boca SM, Carter H, Samayoa J, Bettegowda C, et al. The genetic landscape of the childhood cancer medulloblastoma. Science. DOI: 10.1126/science.1198056

Examples

Run this code

data(ParsonsMB11)
set.seed(188310)
cma.fdr.out <- cma.fdr(cma.alter = GeneAlterMB,
                       cma.cov = GeneCovMB,
                       cma.samp = GeneSampMB,
                       allgenes = TRUE,
                       estimate.p0=FALSE,
                       eliminate.noval=FALSE,
                       filter.mutations=0,
                       M = 2)
names(cma.fdr.out)