Participate in a Follow-up Replication Study of Genetic Patterns in PANS/PANDAS

[bobh]

It has been a long road, but I am in the "home stretch" of data collection for this study: I need only 10 more participants to reach the goal of 70 data sets. So, this is an appeal to any of you that have 23andMe data for your PANS or PANDAS child, to consider participating.

Also, I am taking the opportunity to link to a good podcast on what has been called the "replication crisis". This genetic study I am doing is partly a replication of an initial result in a pilot study. Successful replication is very important because up until recently, half or more of all peer-review, published studies didn't replicate (which implies, they are wrong in their conclusions).

Here is a link to a written transcript of the podcast:
https://www.cbc.ca/radio/ideas/psychologists-confront-impossible-finding-triggering-a-revolution-in-the-field-1.5344467

and here is the actual podcast itself if anyone interested prefers listening:
http://21393.mc.tritondigital.com/CBC_IDEAS_P/media-session/1cd5524d-881b-45af-b51a-0957b6046af8/ideas-vUWHypta-20191101.mp3
(right click and "save as" to download)

[bobh]

I have received the 70th data set (the goal for this replication study), and of course I couldn't resist checking the results.

 

I was very surprised to find that the main SNP we were trying to replicate (the one related to MS that was unusually common in the pilot study) was not in fact unusual among this newest 70 PANDAS/PANS patients. I will be doing some checks around this, but it should be concluded that that SNP is not associated with PANS or PANDAS.

 

But another SNP, rs2811178 (on the DPYD gene, 97981242T>C) has been shown in this study to have met criteria for statistical significance. That SNP's T allele (which normally has a frequency of about 38% among the mostly European population of the participants) is more common allele than the main allele in the P/P population for which we have data. It is not an "if and only if" genetic variation. Most certainly not everyone with a T allele has PANS or PANDAS (because a lot of the general population has at least 1 T), and there are also plenty with P/P (18% among our data) that do not have a T at all for rs2811178. It is just a risk factor for P/P, that'ss all.

 

The above is a good find, but overall, the result from this study is disappointing for me. Besides a replication, I was hoping for more than just one out of the other 9 we were testing for this study to turn out to be significant. One or 2 of these might have been significant with more data (because they were still a little bit unusual), but I won't be trying to get more data to show that, or to do a replication of this rs2811178 finding.

 

This poor result forces me back to the drawing board, though. I am considering checking the data I have now amassed between the pilot study and this one, for all SNPs called by 23andme. Besides being a tremendous amount of work, there is a significant "penalty" in the analysis for checking so many SNPs, that makes proving significance quite a bit harder, so I will think more about this before embarking on it.

[bobh]

We did find one other statistically significant SNP when combining data from the previous pilot study and this replication study: rs1042098. The tallies for both significant SNPs are:

 

                    AA AG GG

rs1042098 66 28 5    (there were 99 calls)

expected# 42 45 12

 

                    TT CT CC

rs2811178 23 21 10   (there were 54 calls)

expected#    8 25 21

 

Although there are 138 participants (pilot + replication studies combined), not all versions of 23andMe and AncestryDNA called both SNPs above. For example, neither of the two SNPs above are called in 23andMe v5, nor most AncestryDNA data sets.

Edited December 21, 2019 by bobh
claried that results were "statistically significant"

[bobh]

If anyone is interested in participating in this project by contributing data, I will be taking extra data until the end of this month (Jan/20).  This time, I will be taking AncestryDNA data too.

I have decided to go ahead with a 3rd (and final) phase of this genetic association study, where I check all the SNPs called in 23andMe and AncestryDNA, not just a small, predetermined list. Well, not quite all, because some aren't reported on in databases, or don't have an rsID, but it is huge chunk of work. For those interested in technical details, the plan for this study can be found here: https://osf.io/q4y2k/registrations.