Question about batch effects in refine.bio datasets

Question

Question about batch effects in refine.bio datasets

kengcher opened this issue 3 years ago · comments

Hi!

I am trying to understand whether batch effects are corrected for in the refine.bio pipeline.

I downloaded the dataset GSE99039 (microarray) from refine.bio then looked at the dataset using PCA. I noticed that the dataset from refine.bio seem to have a clear separation that does not match any of the metadata.

refine.bio PCA

Hence would like to ask about
i. where is the part in the pipeline that does the (quantile?) normalization
ii. i understand that for the normalized data pipeline if any batch correction was performed.

Thank you.

Jaclyn Taroni · Answer 1 · Mon Mar 29 2021 23:40:29 GMT+0800 (China Standard Time)

Hi @kengcher,

Thanks for your questions and for using refine.bio. The dataset you mention (GSE99039) is submitter-processed, which means we were unable to process the data from raw files and use whatever values the authors submitted to GEO (in this case, it is reported to be RMA normalized values). We do quantile normalize submitter-processed data for delivery, but have less control over what happens prior to that step. We do not perform any batch correction (e.g., ComBat).

Looking at the description for this particular experiment, I would want to know if that separation corresponds to idiopathic PD vs. controls, but you do mention that the separation does not match any of the metadata in your post.

Hope this helps! Let me know if you have additional questions.

Deleted user · Answer 2 · Tue Mar 30 2021 00:10:59 GMT+0800 (China Standard Time)

Hi Jacyln Thanks for getting back! GEO indicates that the CEL files are available: https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE99039 How does refine.bio decide whether to process from CEL or otherwise? The two clusters do not match controls vs diseases: [image: image.png]

…

On Mon, Mar 29, 2021 at 11:40 AM Jaclyn Taroni ***@***.***> wrote: Hi @kengcher <https://github.com/kengcher>, Thanks for your questions and for using refine.bio. The dataset you mention (GSE99039) is submitter-processed, which means we were unable to process the data from raw files and use whatever values the authors submitted to GEO (in this case, it is reported to be RMA normalized values). We do quantile normalize submitter-processed data for delivery, but have less control over what happens prior to that step. We do not perform any batch correction (e.g., ComBat). Looking at the description for this particular experiment, I would want to know if that separation corresponds to idiopathic PD vs. controls, but you do mention that the separation does not match any of the metadata in your post. Hope this helps! Let me know if you have additional questions. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#455 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAIBKGAAXJ3ZW2K5BHCHOLTTGCNP7ANCNFSM4Z33T5XQ> .

Jaclyn Taroni · Answer 3 · Wed Mar 31 2021 00:42:55 GMT+0800 (China Standard Time)

We've looked into why this particular experiment was not processed from raw and believe we may have identified a fix, which we will now need to test. If the fix works, we can expect to make the version of this experiment processed from raw within the next few weeks. We're in the middle of some infrastructure changes for the project, so we appreciate your patience!