Hi, I have a quick question regarding the parametric and non-parametric adjustments in ComBat. I've used some normalised expression data (normalised via RMA) and adjusted for batch effects using parametric approach on two different datasets and I get the attached plots, are these good enough or do I need to run non-parametric adjustments in your opinion? Apart from this bit on the website "If the red and black lines don't match up well, use nonparametric adjustments (par.prior=F)" I couldn't find much else about when it's OK to stick with parametric adjustments or to what level do the lines have to deviate to use non-parametric. Also I tried doing non parametric adjustments and it's taking several hours for 50k probes and still not finished even though took a few minutes to do parametric ones, any idea roughly how long it can take my sample size are 75 samples and approximately 100 for the other dataset for 50k probes. Your valuable opinion will be much appreciated! Many Thanks, Afshan

Afshan, Your plots look fine, go ahead and move forward with the parametric ComBat. In the plots you are looking for gross violations of normality for the batch effects (e.g. heavy skewness or bimodality). If the distributions are somewhat symmetric you won't see much difference between the parametric and non-parametric adjustments in practice. Thanks! Evan On May 12, 2014, at 12:57 PM, Siddiq, Afshan <a.siddiq@imperial.ac.uk<mailto:a.siddiq@imperial.ac.uk>> wrote: Hi, I have a quick question regarding the parametric and non-parametric adjustments in ComBat. I’ve used some normalised expression data (normalised via RMA) and adjusted for batch effects using parametric approach on two different datasets and I get the attached plots, are these good enough or do I need to run non-parametric adjustments in your opinion? Apart from this bit on the website “If the red and black lines don’t match up well, use nonparametric adjustments (par.prior=F)” I couldn’t find much else about when it’s OK to stick with parametric adjustments or to what level do the lines have to deviate to use non-parametric. Also I tried doing non parametric adjustments and it’s taking several hours for 50k probes and still not finished even though took a few minutes to do parametric ones, any idea roughly how long it can take my sample size are 75 samples and approximately 100 for the other dataset for 50k probes. Your valuable opinion will be much appreciated! Many Thanks, Afshan <lcm_combat_plots.jpeg><od_combat_plots.jpeg> [[alternative HTML version deleted]]