Perform quality control to ensure that the supplied data set is suitable for Analysis of Partial Variance (APV) within anota.

Share:

Description

Generates a distribution of interaction p-values which are compared to the expected NULL distribution. Also assesses the frequency of highly influential data points using dfbetas for the regression slope and compares the dfbetas to randomly generated simulation data. Calculates omnibus class effects.

Usage

1
2
3
4
anotaPerformQc(dataT=NULL, dataP=NULL, phenoVec=NULL,
generatePlot=FALSE, file="ANOTA_Total_vs_Polysomal_regressions.pdf",
nReg=200, correctionMethod="BH", useDfb=TRUE, useDfbSim=TRUE,
nDfbSimData=2000, useRVM=TRUE, onlyGroup=FALSE, useProgBar=TRUE)

Arguments

dataT

A matrix with cytosolic mRNA data. Non numerical rownames are needed.

dataP

A matrix with translational activity data. Non numerical rownames are needed.

phenoVec

A vector describing the sample classes (each class should have a unique identifier). Note that dataT, dataP and phenoVec must have the same sample order so that column 1 in dataP is the translational activity data for a sample, column 1 in dataT is the cytosolic mRNA data and position 1 in phenoVec describes the sample class.

generatePlot

anota can plot the regression for each gene. However, as there are many genes, this output is normally not informative. Default is FALSE, no individual plotting.

file

If generatePlot is set to TRUE use file to set desired file name (prints to current directory as a pdf). Default is "ANOTA_Total_vs_Polysomal_regressions.pdf"

nReg

If generatePlot is set to TRUE, nReg can be used to limit the number of output plots. Default is 200.

correctionMethod

anota adjusts the omnibus interaction and sample class p-values for multiple testing. Correction method can be "Bonferroni", "Holm", "Hochberg", "SidakSS", "SidakSD", "BH", "BY", "ABH" or "TSBH" as implemented in the multtest package or "qvalue" as implemented in the qvalue package. Default is "BH".

useDfb

Should anota assess the occurrence of highly influential data points (defult is TRUE)?

useDfbSim

The random occurrence of dfbetas can be simulated. Default is TRUE. FALSE represses simulation which reduces computation time but makes interpretation of the dfbetas difficult.

nDfbSimData

If useDfbSim is TRUE the user can select the number of samplings that will be performed per step (10 steps with different correlations between the translationally activty and the cytosolic mRNA level). Default is 2000.

useRVM

The Random Variance Model (RVM) can be used for the omnibus sample class comparison. In this case the effect of RVM on the distribution of the interaction significances needs to be tested as well. Default (TRUE) leads to calculation of RVM p-values for both omnibus interactions and omnibus sample class effects.

onlyGroup

It is possible to suppress the omnibus interaction analysis and only perform the omnibus sample class effect analysis. Default is FALSE (analyse both interactions and sample class effects.)

useProgBar

Should the progress bar be shown. Default is TRUE, show progress bar.

Details

The anotaPerformQc performs the basic quality control of the data set. Two levels of quality control are assessed, both of which need to show good performance for valid application of anota. First, anota assumes that there are no interactions (for slopes). The output for this analysis is both a density plot and a histogram plot of both the raw p-values and the p-values adjusted by the selected multiple correction method (if RVM was used, the second page shows the same presentation using RMV p-values). anota requires a uniform distribution of the raw interaction p-values for valid analysis of differential translation. anota also assesses if there are more data points with high influence on the regression analyses than would be expected by chance. anota identifies influential data points as data points that influence the slope of the regression using standardized dfbeta (dfbetas). In the literature there are multiple suggestions of what should be regarded as an outlier dfbetas (dfbetas>1, dfbetas>2, dfbetas>3, dfbetas>(2/sqrt(N)), dfbetas>(3/sqrt(N)), dfbetas>(3.5*IQR)). Independent of which threshold is preferred, what is of interest is the comparison to the underlying distribution. As this distribution is unknown, we simulate random data sets assuming that the cytosolic mRNA level and the translationally active mRNA levels are normally distributed and that there is a correlation between the cytosolic and the translationally active mRNA level. Following such simulation the frequencies of outlier dfbetas (using all thresholds) is compared to the frequencies found in the simulated data set. The function also performs an omnibus sample class effect test if there are more than 2 sample classes. It is possible to use RVM for the omnibus sample class statistics. If RVM is used, it is necessary to verify that the interaction RVM p-values also follow the expected NULL distribution. A rare error can occur when data within dataT or dataP from any gene and any sample class has no variance. This is reported as "ANOVA F-TEST on essentially perfect fit...". In this case those genes that show no variance for a sample class within either dataT or dataP need to be removed before analysis. Trying a different normalization method may fix the problem.

Value

anotaPerformQc generates several graphical outputs. One output ("ANOTA_interaction_p_distribution.pdf") shows the distribution of p-values and adjusted p-values for the omnibus interaction (both using densities and histograms). The second page of the pdf displays the same plots but for the RVM statistics if RVM is used. One output ("ANOTA_simulated_vs_obtained_dfbs.pdf") shows bar graphs of the frequencies of outlier dfbetas using different dfbetas thresholds. If the simulation was enabled (recommended) these are compared to the frequencies from the random data set. One optional graphical output shows the gene by gene regressions with the sample classes indicated. In the case where RVM is used, a Q-Q plot and a comparison of the CDF of the variances to the theoretical CDF of the F-distribution is generated (output as "ANOTA_rvm_fit_for_....jpg") for both the omnibus sample class and the omnibus interaction test. The function also outputs a list object containing the following data:

omniIntStats

A matrix with a summary of the statistics from the omnibus interaction analysis containing the following columns: "intMS" (the mean square for the interaction); "intDf" (the degrees of freedom for the interaction); "residMS" (the residual error mean square); "residDf" (the degrees of freedom for the residual error); "residMSRvm" (the mean square for the residual error after applying RVM); "residDfRvm"(the degrees of freedom for the residual error after applying RVM); "intRvmFval" (the F-value for the RVM statistics); "intP" (the p-value for the interaction); "intRvmP" (the p-value for the interaction using RVM statistics); "intPAdj" (the adjusted [for multiple testing using the selected multiple testing correction method] p-value of the interaction); "intRvmPAdj"(the adjusted [for multiple testing using the selected multiple testing correction method] p-value of the interaction using RVM statistics).

omniGroupStats

A matrix with a summary of the statistics from the omnibus sample class analysis containing the following columns:"groupSlope" (the common slope used in APV); "groupSlopeP" (if the slope is <0 or >1 a p-value for the slope being <0 or >1 is calculated; if the slope is >=0 & <=1 this value is set to 1); "groupMS" (the mean square for sample classes); "groupDf" (the degrees of freedom for the sample classes); "groupResidMS" (the residual error mean square); "groupResidDf" (the degrees of freedom for the residual error); "residMSRvm" (the mean square for the residual error after applying RVM); "groupResidDfRvm"(the degrees of freedom for the residual error after applying RVM); "groupRvmFval" (the F-value for the RVM statistics); "groupP" (the p-value for the sample class effect); "groupRvmP" (the p-value for the sample class effect using RVM statistics); "groupPAdj" (the adjusted [for multiple testing using the selected multiple testing correction method] p-value of the sample class effect); "groupRvmPAdj"(the adjusted [for multiple testing using the selected multiple testing correction method] p-value of the sample class effect using RVM statistics).

correctionMethod

The multiple testing correction method used to adjust the nominal p-values.

dsfSummary

A vector with the obtained frequencies of outlier dfbetas without the interaction term in the model.

dfbetas

A matrix with the dfbetas from the model without the interaction term in the model.

residuals

The residuals from the regressions without the interaction term in the model.

fittedValues

A matrix with the fitted values from the regressions without the interaction term in the model.

phenoClasses

The sample classes used in the analysis. The sample class order can be used to create the contrast matrix when identifying differential translation using anotaGetSigGenes.

sampleNames

A vector with the sample names (taken from the translationally active samples).

abParametersInt

The ab parameters for the inverse gamma fit for the interactions within RVM.

abParametersGroup

The ab parameters for the inverse gamma fit for sample classes within RVM.

Author(s)

Ola Larsson ola.larsson@ki.se, Nahum Sonenberg nahum.sonenberg@mcgill.ca, Robert Nadon robert.nadon@mcgill.ca

See Also

anotaResidOutlierTest, anotaGetSigGenes,anotaPlotSigGenes

Examples

1
 ## See example for \code{\link{anotaPlotSigGenes}}