Batch effect in 16S data

[Larbi Bedrani]

Hi qiimer,
I am trying to compare 16S sequencing data from several studies. Due to the difference in the sequencing technology (illumina vs 454), amplified fragment(s) and extraction kits...etc. the batch effect is clear, basically samples  from each study form an independent "cluster" in a PCoA.

What I did basically is to analyze each study alone using the classical qiime pipeline.  I went naturally for  a close reference OTU picking since the sequenced regions are not the same for every study. Finally I combined the resulting OTU tables together than proceeded to alpha and beta diversity analysis.  

Do you have any recommendations (or references in literature) on how to address the problem of sequencing parameters, or on how to correct for the batch effect? I thought that maybe the length of the generated sequences may differ from a technology to the other and maybe a trimming of the sequences is needed. If that's the case, is there any script on qiime to do this?

Thanks.
Larbi

[Kyle Bittinger]

Larbi,

You have a difficult task, because any of the three factors you mention could introduce a strong batch effect, and you seem to have no way of knowing which is responsible for the differences you see.  This makes the problem difficult to fix.

The different error profiles in 454 and Illumina data make data from the two platforms difficult to merge.  The length difference is not your biggest issue, rather it is the insertion/deletion error rate in the 454 data is much higher than in the Illumina data.

If the platforms are the same then you've got a shot with stuff like sequence trimming, chimera removal, and reference-based OTU selection.  Let me know more details if you're still having trouble with data from the same platform.

[Larbi Bedrani]

Thanks Kyle for the clear answer. I will let you know...
Larbi