diffexp.lite: diffexp.lite

In kuppal2/xmsPANDA: R Package for Biomarker Discovery, Network, and Data Exploratory Analysis

Description

A "lite" version of the diffexp() function with key features for data preprocessing, feature selection, PCA/HCA, and correlation analysis

Usage

diffexp.lite(Xmat = NA, Ymat = NA, outloc = NA, summary.na.replacement = "halffeaturemin", missing.val = 0, all.missing.thresh = 0.1, group.missing.thresh = 0.7, input.intensity.scale = "raw", normalization.method = c("log2transform", "znormtransform", "lowess_norm", "log2quantilenorm", "quantile_norm", "rangescaling", "paretoscaling", "mstus", "eigenms_norm", "vsn_norm", "sva_norm", "tic_norm", "cubicspline_norm", "mad_norm", "none"), rsd.filt.list = 1, pca.global.eval = TRUE, pairedanalysis = FALSE, featselmethod = c("limma"), pvalue.thresh = 0.05, fdrthresh = 0.05, fdrmethod = "BH", foldchangethresh = 0, analysismode = "classification", pls_vip_thresh = 2, num_nodes = 2, optselect = TRUE, max_comp_sel = 1, hca_type = "two-way", output.device.type = "png", timeseries.lineplots = FALSE, alphabetical.order = FALSE, ylab_text = "Abundance", boxplot.type = "ggplot", color.palette = c("journal", "npg", "nejm", "jco", "lancet", "custom1", "brewer.RdYlBu", "brewer.RdBu", "brewer.PuOr", "brewer.PRGn", "brewer.PiYG", "brewer.BrBG", "brewer.Set2", "brewer.Paired", "brewer.Dark2", "brewer.YlGnBu", "brewer.YlGn", "brewer.YlOrRd", "brewer.YlOrBr", "brewer.PuBuGn", "brewer.PuRd", "brewer.PuBu", "brewer.OrRd", "brewer.GnBu", "brewer.BuPu", "brewer.BuGn", "brewer.blues", "black", "grey65", "terrain", "rainbow", "heat", "topo"), generate.boxplots = TRUE, hca.cex.legend = 0.7, lme.modeltype = "lme.RI", globalcor = FALSE, abs.cor.thresh = 0.4, cor.fdrthresh = NA, net_legend = TRUE, rocclassifier = "svm", limma.contrasts.type = c("contr.sum", "contr.treatment"), limmadecideTests = FALSE, cex.plots = 0.9,hca.labRow.value = TRUE,hca.labCol.value = TRUE,...)

Arguments

Xmat	R object for feature table. If this is given, then feature table can be set to NA.
Ymat	R object for response/class labels matrix. If this is given, then class can be set to NA.
outloc	Provide full path of the folder where you want the results to be written. Eg: C:/My Documents/ProjectA/results/
summary.na.replacement	How should the missing values be represented? Options: "zeros", "halffeaturemin", "halfsamplemin","halfdatamin", "none" "zeros": replaces missing values by 0 "halfsamplemin": replaces missing value by one-half of the lowest signal intensity in the corresponding sample "halfdatamin": replaces missing value by one-half of the lowest signal intensity in the complete dataset "halffeaturemin": replaces missing value by one-half of the lowest signal intensity for the current feature "none": keeps missing values as NAs Users are recommended to perform imputation prior to performing biomarker discovery.
missing.val	How are the missing values represented in the input data? Options: "0" or "NA"
all.missing.thresh	What propotion of total number of samples should have an intensity? Default: 0.5
group.missing.thresh	What propotion of samples in either of the two groups should have an intensity? If at least x for further analysis. Default: 0.7
input.intensity.scale	Are the intensities in the input feature table at raw scale or log2 scale? eg: "raw" or "log2" Default: "raw"
normalization.method	Data transformation and normalization method. Options: "log2transform": log2 transformation "log2quantilenorm": log2 transformation and quantile normalization "znormtransform": auto scaling; each variable will have a mean of 0 and unit variance "quantile_norm": Performs quantile normalization "lowess_norm": Performs lowess normalization "rangescaling": Performs range scaling; scale by the min and max range "paretoscaling": Performs Pareto scaling; scale by the square root of the standard deviation "mstus": MS Total Useful Signal (MSTUS) normalization "sva": Surrogate Variable Analysis (SVA) normalization "eigenms_norm": EigenMS normalization "vsn_norm": variance stabilizing normalization "tic_norm": totial intensity normalization "cubicspline_norm": Cubic spline normalization "mad_norm": Median absolute deviation normalization
pairedanalysis	Is this a paired-study design? TRUE or FALSE If samples are paired, then the feature table and the class labels file should be organized so that the paired samples are arranged in the same order in each group. For example, the first sample in group A and the first sample in group B should be paired.
pca.global.eval	Perform PCA using all variables? TRUE or FALSE
featselmethod	Options: "limma": for one-way ANOVA using LIMMA (mode=classification) "limma2way": for two-way ANOVA using LIMMA (mode=classification) "limma1wayrepeat": for one-way ANOVA repeated measures using LIMMA (mode=classification) "limma2wayrepeat": for two-way ANOVA repeated measures using LIMMA (mode=classification) "lm1wayanova": for one-way ANOVA using linear model (mode=classification) "lm2wayanova": for two-way ANOVA using linear model (mode=classification) "lm1wayanovarepeat": for one-way ANOVA repeated measures using linear model (mode=classification) "lm2wayanovarepeat": for two-way ANOVA repeated measures using linear model (mode=classification) "lmreg": variable selection based on p-values calculated using a linear regression model; allows adjustment for covariates (mode= regression or classification) "logitreg": variable selection based on p-values calculated using a logistic regression model; allows adjustment for covariates (mode= classification) "rfesvm": uses recursive feature elimination SVM algorithm for variable selection; (mode=classification) "RF": for random forest based feature selection (mode= regression or classification) "RFconditional": for conditional random forest based feature selection (mode= regression or classification) "pamr": for prediction analysis for microarrays algorithm based on the nearest shrunken centroid method (mode= classification) "MARS": for multiple adaptive regression splines (MARS) based feature selection (mode= regression or classification) "pls": for partial least squares (PLS) based feature selection (mode= regression or classification) "spls": for sparse partial least squares (PLS) based feature selection (mode= regression or classification) "spls1wayrepeat": for sparse partial least squares (PLS) based feature selection for one-way repeated measures (mode= regression or classification) "spls2wayrepeat": for sparse partial least squares (PLS) based feature selection for two-way repeated measures (mode= regression or classification) "o1pls": for orthogonal partial least squares (OPLS) based feature selection (mode= regression or classification)
pvalue.thresh	p-value threshold. Eg: 0.05^M
fdrthresh	False discovery rate threshold. Eg: 0.05
fdrmethod	Options: "BH", "ST", "Strimmer", "BY","none" "BH": Benjamini-Hochberg (1995) (Default: more conservative than "ST" and "Strimmer") "ST": Storey & Tibshirani (Storey 2001, PNAS) algorithm implemented in the qvalue package "Strimmer": (Strimmer 2008, Bioinformatics) algorithm implemented in the fdrtool package "none": No FDR correction will be performed. fdrthresh will be treated as raw p-value cutoff
foldchangethresh	Secondary feature selection criteria based on fold change threshold. This is performed after statistical significance or importance evaluation.
analysismode	"classification" for group-wise comparison (case vs control) or "regression" for continuous response variables. Default: "classification"
pls_vip_thresh	Threshold for VIP score from PLS/O1PLS. eg: 1
num_nodes	Number of CPU cores to use e.g. 2
optselect	Determine optimal number of PLS components. Default: TRUE
max_comp_sel	Number of PLS components to use for VIP or sparse loading selection (sPLS). Default=1
hca_type	"one-way" or "two-way" HCA
output.device.type	"pdf" or "png"
timeseries.lineplots	Generate lineplots showing longitudinal pattern: TRUE or FALSE Default: FALSE
alphabetical.order	Arrange class labels in alphabetical order versus arranging them based on which class appears first in the class labels file. TRUE or FALSE
ylab_text	Y-axis label in barplots, boxplots, and lineplots Default: "Abundance"
boxplot.type	Type of boxplots: "simple" using the boxplot() function in R or "ggplot" for ggboxplot and geom_boxplot functions
color.palette	Color theme for plots. default: "journal" Options: 1. "journal": color-blind friendly palette 2. built-in R color palettes: "rainbow", "terrain", "heat","topo" 3. RColorBrewer pallettes: "brewer.YlOrRd", "brewer.Purples","brewer.YlGn", "brewer.BuPu","brewer.BuGn","brewer.GnBu", "brewer.YlGnBu", "brewer.RdBu", "brewer.RdYlBu","brewer.PuOr","brewer.PRGn" (color codes: Yl-yellow; Rd-red, Bu-blue, Or-orange, Gn-green, PR-purple) 4. Generate a custom palette by providing colors (e.g. c("orange","blue","green")) Please the color_palettes_xmsPANDA.pdf file on the GitHub page under xmsPANDA/inst
generate.boxplots	Should the boxplots be generated? e.g. TRUE or FALSE
hca.cex.legend	Numeric value indicating the amount by which plotting text and symbols should be scaled relative to the default. e.g 0.7 Set to NA to hide the HCA legend
lme.modeltype	Options for mixed-effects models: RI:Random intercept RIRS: random intercept and random slope models Default: "RI"
globalcor	Perform correlation analysis between selected features and all other features? Options: TRUE or FALSE
abs.cor.thresh	Absolute Pearson correlation coefficient for network analysis. Default: 0.4
cor.fdrthresh	False discovery rate threshold for correlation analysis. Default: 0.05
net_legend	Should the network be displayed for the correlation network? eg: TRUE or FALSE
rocclassifier	Set to NA to turn off k-fold CV classification accuracy and ROC analysis Default: "svm"
limma.contrasts.type	Contrasts method for LIMMA e.g. "contr.sum" for ANOVA like sum contrasts method "contr.treatment" to treat the first group as the reference group and all other groups are compared to the reference
limmadecideTests	Perform decide tests for LIMMA to perform multiple testing and assign up, down, or not significant. TRUE or FALSE.
cex.plots	Relative factor to change font size of text in plots e.g.: 0.8 or 2 Default: 1
hca.labRow.value	Show variable (row) names in hierarchical clustering analysis heatmaps e.g. TRUE or FALSE
hca.labCol.value	Show sample (column) names in hierarchical clustering analysis heatmaps e.g. TRUE or FALSE

Details

The "lite" version requires fewer computational resources. The function performs data transformation, normalization, feature selection, evaluates the predictive accuracy of the FDR significant features using k-fold cross-validation with a Support Vector Machine classifier, performs hierarchical clustering analysis, correlation analysis, and principal component analysis.

Value

diffexp_metabs	Best set of discriminatory features.
all_metabs	Results for all features.
mw.an.fdr	Metabolome-wide significant correlation network of differentially expressed metabolites.
targeted.an.fdr	Correlation network of differentially expressed metabolites with targeted metabolites.

Following files are generated in the parent output location: Manhattan plots: showing metabolome wide p-values; Heatmap from Two-way hierarchical clustering analysis; Pairwise score plots from Principal Component Analysis; PCA score distribution plots; ROC plots; List of differentially expressed metabolites; Boxplots of differentially expressed metabolites; Correlation network figure and matrix; Pairwise correlation matrix CIRCOS format ready to be uploaded to: http://mkweb.bcgsc.ca/tableviewer/visualize/ Or uploaded to Cytoscape gml format

Author(s)

Karan Uppal <[e-mail]>

kuppal2/xmsPANDA documentation built on May 15, 2021, 5:48 a.m.