Version 1.0.5 - "confounded confounders" will now be removed from the status label of a confounded variable. - Example: If the association of D with feature Y is confounded by C1 and C2, but the association of C1 with Y is in turn confounded by C2, then only C2 will be listed as confounder of Y~D - improvments in some logging messages - featureMat is now checked for non-numeric columns at start-up. Version 1.0.4 - MetaDeconfound() - improved error handling for glms/glmers whith highly collinear metavariables - bugfix in confidence interval calculations - bugfix for ncol(fatureMat) == 1 - new optional input argument "mediationMat" - perform mediation analysis between features of featureMat and mediadtionMat, controlled for effects from variables in metaMat - features in mediationMat will not be tested for confounding effects on each other or variables from metaMat - new optinoal argument "mediationMat" - supply a second omics space to perform mediation analysis between featureMat and mediationMat - new optional argument "adjustLevel" - specify different multiple testing p-value correction approaches - 1: correct for number of features per metavariable (default) - 2: correct for number of features x number of meatavariables - 3: correct for number of features x number of mediationMat features (default when mediationMat is supllied) - BuildHeatmap() - new options "starSize" and "starNudge_y" to control size and positioning of confounding status asterisks/circles - metavariables can be split into separate groups by adding a "grounpingVar" columns to metaDeconfOutput. - this is done by default when mediationMat is supplied to MetaDeconfound() - uses facet_grid(), so output can be tweaked using "+ theme(strip.XXX)" options - BuildConfounderMap() - new plot generating function - for each feature it creates a circos plot of confounder-confounded relations between metavariables associated to this feature. - returns a list of ggraph ggplot2 plots - ggraph package must be loaded for plotting elements of the returned list. Version 1.0.3 - implementation of new funtion GetPartialEfSizes() - adds partial R-squared as additional effect size metric - new output column "partial": additional variance of a feature explained by adding a metavariable to a linear (mixed effects) model containing all other metavariables associated to the feature - new output column "partialRel": proportion of explainable variance of a feature tracable to a metavariable (e.g. partialRel = 0.5 --> metavariable x can explain 50% of all explainable variance of fature y -- all other metavariables significantly associated to feature y can explain the remaining 50% of overall explainable variance of fature y) - new output column "PartialNorm": proportion of explained variance of a metavariable set in relation to maximally explainable variance (i.e. variance not explainable by all other metavariables) - bug fix of bug disabling computation of confidence intervalls of lrts - implementaion of truly linear computation mode when nnodes = 1 - additional test to filter out non trustworthy lrts in some scenarios with highly colinear metavariables - stricter tests for complete separaion when glms are computed - BuildHeatmap() - added "plotPartial" argument to choose between - "marginal": plotting naive/marginal effect sizes based on CLiff's Delta/Spearman's Rho - "partial": plotting signed partial R-squared, including an additional column showing the total explainable variance per feature - "partialRel": plotting proportion of explainable variance of a feature tracable to a metavariable - "partialNorm": plotting respective normalized partial effect size Version 1.0.2 - CRAN release Version 1.0.1 - fixes for CRAN resubmission Version 1.0.0 - CRAN submission Version 0.3.1 - reduced side effects after interrupted metadeconfoundR runs (tmp file removal) - bugfix: Qcutoff now correctly used to collect potential confounders from naive associations - improved initial input data quality control (order, problematic row-/colnames, class) - minor bugfixes in Buildheatmap() - added LogLevel argument to change verbositiy of logging Version 0.3.0 - randomVar argument is now simplified: just supply character vector of metaMat variables to be treated as random effects (e.g. randomVar = c("var1", "var5")) - NEW fixedVar argument implemented: same usage as randomVar, but adds fixed effect terms to all models - implemented computational speed-up for datasets with large number of metavariables - metadeconfoundR is now compatible with Unix AND Windows operation systems - fixed a bug that lead to an error when a feature with extremely low number non-NA values was present in featureMat - improved error messaging for an error in BuildHeatmap function - addition of NEW "rawCount" mode, that runs modelling steps on not normalized/rarefied data by including totalReadCount per sample information into glm/glmer comparisons - for naive significance test and effect size calculation, rawCounts will be normalized by dividing by totalReadCount per sample - updated implementation of logistic regression mode: test for association of binary features to metavariables by setting "logistic = TRUE" - experimental NEW feature: return all computed models alongside normal Metadeconfound() output by setting collectMods = TRUE. WARNING: For now, this only works without utilizing parallel processing in the model building step. Larger datasets might take a lot of time processing. - NEW BuildHeatmap() argument "tileBordCol" sets tile border color (Default "black") - NEW BuildHeatmap() argument "reOrder" to turn on/of sorting of features and metavariables in the reulsting plot Version 0.2.8 - changes in status label names and corresponding plotting behavior - reversed default color scheme for effect size plotting (red = low, blue= high) - computation of confidence intervals in model building step now default (with added OK_d label for deconfounded but doubtful associations) Version 0.2.7 - improved error messages and behavior for no/low number of significant associations - can now plot results with only a single feature/and or metadata Version 0.2.6 - fixed bug in legend creation of BuildHeatmap(cuneiform = TRUE), direction of association now labeled correctly Version 0.2.5 - fixed bug in "keepFeature" functionality of BuildHeatmap() - fixed bug in BuildHeatmap() that resulted in incomplete removal of features showing not a single significant association - implemented speedup for runs of Metadeconfound(startStop = "naiveStop") with large number of metadata Version 0.2.4 - Buildheatmap(cuneiform = TRUE) can now handle NA or zero effect sizes (added NEW symbol) Version 0.2.3 - fixed bug that only occurred when metadata only has one column - implemented speedup when startStop = "naiveStop" Version 0.2.2 - added functionality to BuildHeatmap() - range of shown effectsizes in using d_range parameter (set to "full" for consistent legend between plots) - range of colors used to show effect sizes can be changed using d_col - matavariables and features, that should be kept in the plot even without passing q_cutoff and d_cutoff cutoffs, can be supplied using keepMeta and keepFeature parameters Version 0.2.1 - added functionality to Metadeconfound() - output can now also be generated in long format - added functionality to BuildHeatmap() - input can now also be read in long format - "human readable"" names for features and metadata can be supplied in addition to the Metadeconfound() output and will be plotted instead of "machine readable" names - slightly improved help pages Version 0.2.0 - NEW function ImportLongPrior() to easily import prior knowledge of feature metadata associations present in the current dataset. - slightly improved help pages Version 0.1.9 - bugfix concerning random effect variable behavior - minor aesthetic improvements in BuildHeatmap() output Version 0.1.8 - introduction of global parameter "logistic" - logistic = TRUE: analyzing binary features instead of continuous using logistic regression models Version 0.1.7 - bug fix in BuildHeatmap function Version 0.1.2 - critical bug fixed, that resulted in greatly increased "NS" status labeling