What Is A Correct Way To Assess Batch Effect?

Question

[vanzetti]

Hello.

I have some proteomic data, where I discovered batch effect by clustering the experiments using euclidean distance hierarchical clustering between columns (it clustered biological repeats together in separate groups). Use of ComBat removed most of the effect. However, what I'm asking is, was the clustering I did the best way to look for batch effect?

Thanks

[Simon Cockell]

Your biological replicates clustered together and you considered this a batch effect? Surely unwanted batch effects would be indicated by your samples clustering out according to some criteria other than biological replicate??

[Damian Kao]

His biological samples clustered together in separate groups apparently. I am guessing he meant if there were 4 biological replicates, all 4 clusters together, but there are distinct subgroups within the replicates that also cluster.

The question is how heterogenous are the subgroups he considers them to be due to batch effects.

[vanzetti]

Exactly. I'm doing a time course, and expect clustering by time-points. And after using ComBat.R, they were mostly clustered by time-points.

[Simon Cockell]

Gotcha, apologies for confusion.