ComBat (SVA) Application - Two Technical Effects
Entering edit mode
8.8 years ago


I'm basically trying to follow up a post I found -!topic/combat-user-forum/PcTxNlaUmAI

Ultimately, I'm trying to perform a batch correction on Microarray data with two known, and different technical effects (batches).

I realise this can be accounted for using an additive model, but I'm interested on how the ComBat function in SVA could be used to tackle such a problem.

I'm wondering which option is the "done thing":

Apply combat twice - Apply once on Technical effect A, then apply resultant expression set on Technical effect B

Merge All Technical Effects - Regardless of origin, merge Technical Effect A and B into one. i.e, Technical Effect A (two batches; 1 and 2), Technical Effect B (two batches; 1 and 2) - Inform ComBat of 4 Batches.

Any suggestions?


P.S. I originally posted this to Bioconductor Support, but to no avail yet. In order to avoid the loss of information in cross posting, I'll update both posts with the relevant responses, or delete one of the posts (depending on what the mods prefer).

ComBat sva bioconductor • 4.6k views
Entering edit mode

Here is my earlier post on Batch Correction: A: removing batch effects using ComBat and SVA

I would recommend you to combine all the dataset you have and then get the batch information using sva function. Now sva gives you surrogate variables (which is batch plus information about other latent variables). Plot surrogate variables vs sample number, check how many clusters are present. Your cluster will give you a rough idea about batches in the dataset. Then provide this batch information to ComBat and do a sanity check using PCA if the samples (with same background or genotype) are clustering together or not.

Entering edit mode

Hi, thanks for your reply - This post is from nearly 2 years ago! Nowadays, I would stay away from ComBat, due to it's tendency to exaggerate effects due to the model you give it (see here), which basically means that when you give it a model, you're telling it what experimental groups you expect, and the effect of those groups is increased. Generally speaking, I stick with incorporating the batch into the linear model design, and if I need to "remove" the effect in an expression matrix, then I make use of Limma's RemoveBatchEffect function

Entering edit mode

I am sorry but for some reason the post came up on the "Biostars Latest question". I think it was because the Biostar modified your post. However, thanks for sending the paper. I will go through it.


Login before adding your answer.

Traffic: 806 users visited in the last hour
Help About
Access RSS

Use of this site constitutes acceptance of our User Agreement and Privacy Policy.

Powered by the version 2.3.6