clippda-package: A package for clinical proteomics profiling data analysis
In clippda: A package for the clinical proteomic profiling data analysis

Description Details Author(s) References Examples

This package is still under development but it is intended to provide a range of tools for analysing clinical genomics, methylation and proteomics, data with the non-standard repeated expression measurments arising from technical replicates. Most of these studies are observational case-control by design and the results of analyses must be appropriately adjusted for confounding factors and imbalances in the data. This regression-type problem is different from the regression problem in limma, in which all the covariates are some kind of contrasts and are therefore important. Our method is specifically suitable for analysing single-channel microarrays and proteomics data with repeated probe, or peak measurements, especially in the case where there is no one-to-one correspondence between cases and controls and the data cannot be analysed as log-ratios. In the current version (version 0.1.0), we are more concerned with the problem of sample size calculations for these data sets. But some tools for pre-processing of the repeated peaks data, including tools for checking for the consistency in the number of replicates across samples, the consistency of the peak information between replicate spectra and tools for data formatting and averaging, are included. clippda also implements a routine for evaluating differential-expression between cases and controls, especially for data in which each sample is assayed more than once, and are obtained from studies which are observational, or those for which the data are heterogeneous (e.g. data for cancer studies in which controls are not directly sampled, but are obtained from samples from suspected cases that turn out to be benign disease, after an operation, for example. In this case there could be serious imbalances in demographics between the cases and controls). The test statistics considered are derived from the methods developed by Nyangoma et al. (2009). These new methods for evaluating differential-expression are compared with the empirical Bayes method in the limma package. To limit the number of false positive discoveries, we control the tail probability of the proportion of false positives, (TPPFP). Further details can be found in the package vignette.

Package:	anRpackage
Type:	Package
Version:	0.1.0
Date:	2009-03-25
License:	GPL (>=2)
LazyLoad:	yes

This package provides a method for calculating the sample size required when planning proteomic profiling studies using repeated peak measurements. At the planning stage, an experimenter typically does not yet have information on the heterogeneity of the data expected. We provide a method which makes it possible to input, and adjust for the effect of, the expected heterogeneity in the sample size calculations. The code for calculating sample size is the sampleSize function. It requires the computation of the between- and within-sample variations, the differences in mean between cases and controls, the intraclass correlations between duplicate peak data, and the heterogeneity correction factor. These can be computed from pilot data using the functions: betweensampleVariance, withinsampleVariance, replicateCorrelations and FisherInformation, respectively. Before the data can be analysed using these functions, it must be adequately pre-processed, and this package provides a number of tools for doing this. We provide a grid of the clinically important differences versus protein variances (with superimposed sample size contours). On this grid, we have plotted sample sizes computed using parameters from several real-life proteomic data from a range of cancer-types, fluid-types, cancer stages and experimental protocols of SELDI and MALDI. These values provide sample size ranges which may be used to estimate the number of samples required. The example below takes you through some of the processes of sample size calculations using this package.

Stephen Nyangoma

Maintainer: S Nyangoma s.o.nyangoma@bham.ac.uk

Birkner, et al. Issues of processing and multiple testing of SELDI-TOF MS Proteomic Data. Stat Appl Genet Mol Biol 2006, 5.1

Nyangoma, Stephen O.; Collins, Stuart I.; Altman, Douglas G.; Johnson, Philip; and Billingham, Lucinda J. (2012) Sample Size Calculations for Designing Clinical Proteomic Profiling Studies Using Mass Spectrometry, Statistical Applications in Genetics and Molecular Biology: Vol. 11: Iss. 3, Article 2.

Nyangoma SO, et al. Multiple Testing Issues in Discriminating Compound-Related Peaks and Chromatograms from High Frequency Noise, Spikes and Solvent-Based Noise in LC - MS Data Sets. Stat Appl Genet Mol Biol 2007, 6, 1, Article 23

Smyth GK, et al.: Use of within-array replicate spots for assessing differential expression in microarray experiments. Bioinformatics 2005, 21, 2067 - 75

Smyth GK: Linear models and emperical Bayes methods for assessing differential expression in microarray experiments. Stat Appl Genet Mol Biol 2004, 3, 1, Article 3

#########################################################################################
# The routine for calculating sample size required when planning a clinical proteomic
# profiling study is provided in the sampleSize function. First, this functions performs 
# computations for sample size parameters, that include: the biological variance, the
# tecnical variance, the differences to be estimated, the intraclass correlation 
# (if unknown). These computations are done ase follows: 
#########################################################################################

#########################################################################################
# biological variation, difference to be estimated, and the p-values for differential-
# expression are computed using the generic function: betweensampleVariance
# It requires data of a aclinicalProteomicsData class, as input 
########################################################################

###################################################
# Creating data of a aclinicalProteomicsData class
###################################################

data(liverdata)

data(liver_pheno)

OBJECT=new("aclinicalProteomicsData")

OBJECT@rawSELDIdata=as.matrix(liverdata)

OBJECT@covariates=c("tumor" , "sex")

OBJECT@phenotypicData=as.matrix(liver_pheno)

OBJECT@variableClass=c('numeric','factor','factor')

OBJECT@no.peaks=53

Data=OBJECT

#################################################################################
# Data manipulation carried out internally by the betweensampleVariance function 
#################################################################################

rawData <- proteomicsExprsData(Data)

no.peaks <- Data@no.peaks

JUNK_DATA <- sampleClusteredData(rawData,no.peaks)

JUNK_DATA=negativeIntensitiesCorrection(JUNK_DATA)

# we use the log base 2 expression values

LOG_DATA <- log2(JUNK_DATA)

#######################################################################################
# compute biological variation, difference to be estimated, and the p-values 
#######################################################################################

BiovarDiffSig <- betweensampleVariance(OBJECT)

BiovarDiffSig

#####################
# technical variance
#####################

sample_technicalVariance(Data)

#######################################################################################
# heterogeneity correction-factor is the second diagonal element of the output 
# matrix from the fisherInformation function, i.e. from the expected Fisher Information
########################################################################################

Z <- as.vector(fisherInformation(Data)[2,2])/2
Z

###################################################################################
# The outputs of these functions are converted into statistics used in
# the sample size calculations using a wraper function sampleSizeParameters
# it gives the consensus parameter values. You must specify the p-value and the 
# intraclass correation, cutoff. The description of how these parameters
# are chosen is given in Nyangoma, et al. (2009).  
###################################################################################

intraclasscorr  <-  0.60 #cut-off for intraclass correlation

signifcut <- 0.05      #significance cut-off

sampleparameters=sampleSizeParameters(Data,intraclasscorr,signifcut)

#######################################################################################
# SAMPLE SIZE CALCULATIONS
#The function sampleSize calculates the protein variance, difference to be estimated, 
# the technical varaince. These parameters are computed from statistics of peaks with 
# medium biological variation. 
# It also gives sample sizes for beta=c(0.90,0.80,0.70) and alpha = c(0.001, 0.01,0.05)
#######################################################################################

samplesize <- sampleSize(OBJECT,intraclasscorr,signifcut)
samplesize