# David Causeur

#### 4 packages on CRAN

Functions for signal detection and identification designed for Event-Related Potentials (ERP) data in a linear model framework. The functional F-test proposed in Causeur, Sheu, Perthame, Rufini (2018, submitted) for analysis of variance issues in ERP designs is implemented for signal detection (tests for mean difference among groups of curves in One-way ANOVA designs for example). Once an experimental effect is declared significant, identification of significant intervals is achieved by the multiple testing procedures reviewed and compared in Sheu, Perthame, Lee and Causeur (2016, <DOI:10.1214/15-AOAS888>). Some of the methods gathered in the package are the classical FDR- and FWER-controlling procedures, also available using function p.adjust. The package also implements the Guthrie-Buchwald procedure (Guthrie and Buchwald, 1991 <DOI:10.1111/j.1469-8986.1991.tb00417.x>), which accounts for the auto-correlation among t-tests to control erroneous detection of short intervals. The Adaptive Factor-Adjustment method is an extension of the method described in Causeur, Chu, Hsieh and Sheu (2012, <DOI:10.3758/s13428-012-0230-0>). It assumes a factor model for the correlation among tests and combines adaptively the estimation of the signal and the updating of the dependence modelling (see Sheu et al., 2016, <DOI:10.1214/15-AOAS888> for further details).

The functions provided in the FADA (Factor Adjusted Discriminant Analysis) package aim at performing supervised classification of high-dimensional and correlated profiles. The procedure combines a decorrelation step based on a factor modeling of the dependence among covariates and a classification method. The available methods are Lasso regularized logistic model (see Friedman et al. (2010)), sparse linear discriminant analysis (see Clemmensen et al. (2011)), shrinkage linear and diagonal discriminant analysis (see M. Ahdesmaki et al. (2010)). More methods of classification can be used on the decorrelated data provided by the package FADA.

The method proposed in this package takes into account the impact of dependence on the multiple testing procedures for high-throughput data as proposed by Friguet et al. (2009). The common information shared by all the variables is modeled by a factor analysis structure. The number of factors considered in the model is chosen to reduce the false discoveries variance in multiple tests. The model parameters are estimated thanks to an EM algorithm. Adjusted tests statistics are derived, as well as the associated p-values. The proportion of true null hypotheses (an important parameter when controlling the false discovery rate) is also estimated from the FAMT model. Graphics are proposed to interpret and describe the factors.

Various tools dealing with batch effects, in particular enabling the removal of discrepancies between training and test sets in prediction scenarios. Moreover, addon quantile normalization and addon RMA normalization (Kostka & Spang, 2008) is implemented to enable integrating the quantile normalization step into prediction rules. The following batch effect removal methods are implemented: FAbatch, ComBat, (f)SVA, mean-centering, standardization, Ratio-A and Ratio-G. For each of these we provide an additional function which enables a posteriori ('addon') batch effect removal in independent batches ('test data'). Here, the (already batch effect adjusted) training data is not altered. For evaluating the success of batch effect adjustment several metrics are provided. Moreover, the package implements a plot for the visualization of batch effects using principal component analysis. The main functions of the package for batch effect adjustment are ba() and baaddon() which enable batch effect removal and addon batch effect removal, respectively, with one of the seven methods mentioned above. Another important function here is bametric() which is a wrapper function for all implemented methods for evaluating the success of batch effect removal. For (addon) quantile normalization and (addon) RMA normalization the functions qunormtrain(), qunormaddon(), rmatrain() and rmaaddon() can be used.