Thursday, October 11, 2018

New Depression Study Finding 102 Variants. What is Replication?

A new Depression study claiming 102 genetic variants has just come out (pre-publish).  I don't want to do an extended critique, so I will stick to a few main points:

1. For starters, I keep seeing this statement:  "...however depression is a polygenic trait influenced by many genetic variants each of small effect," being presented as factual rather than theoretical.  It is an unproven assumption based on the failure to find specific genes or genetic variants of larger effect.  I don't think it's splitting hairs to point this out.
2. The claimed "replication" of the this study with a large 23andMe database is based on some kind of mathematical probability comparison.  There has not been a GWAS performed on the 23andMe database with a direct comparison of significant genetic variants as the term replication might suggest.  Replication, as I understand it, would be matching of significant variants from GWAS's of two independent studies.  I look forward to the results of any such independent GWAS study of the 23andMe dataset with a comparison to the results of this one for common significant genetic variants (However, I'm predicting that rather than doing that, they will take this current study and combine it with that one, in which case we won't be able to do such a comparison).
3. The three studies used in this meta-analysis had some overlap in their datasets, such that the previous Howard Study and the Wray study had some common loci (14 out of 44), so I believe that there were 62 unique variants between the studies of the total of 76 noted at face (15, 17 and 44 from the three studies). Because of the continuing meta-analysis building from one study to another, we can't say whether there were any replicated loci between the studies unless we were to assess their data independently (which would likely change the results).  One would assume, however, that most or all of the 14 matches in the Wray and previous Howard study were due to overlapping datasets, however.  Where I'm going with this is that we have three studies that likely produced completely different results and that are now being combined.
4. About two thirds of the variants noted as significant in any of the previous studies maintained that significance in the meta-analysis.  This result is difficult to assess and one might ask what we can really glean from this when any significant variant found in the meta-analysis, would be bolstered by the dataset that found it to be significant in one of the previous studies.  Moreover, I am not able to determine how many of those that maintained significance in the Wray study were the same as those found in the previous Howard study due to overlap.  In any case, a reasonable question here is why so many variants lose significance with the addition of more data?  One would think that more data would make true significant variants even more significant.
5. Although, as noted above, many of the variants did not maintain significance, there were 40 new variants found (102 - 62 = 40 if I am counting correctly).  Why were these new variants not seen in the previous studies?  It seems that increasing N increases significance for some variants and decreases it for others.
6.  The authors also noted a Han Chinese depression study., but did not include it in the meta-analysis.  In the world of GWAS, are we expecting different depression mechanisms among different races?  Most studies stick to white European, which gives a consistent sample, but I would think that mixing different studies with a different ethnic and geographic population could decrease the overall risk of population stratification (assuming they have the same mechanism).  Anyway, feel free to comment on this.
7. Also not included in the analysis were two previous studies that had not found any depression variants (N = 53,000 between the two).  Not clear why these were left out of the study.  There might be some other Depression studies that were left out, but it is difficult to determine due to the meta-analysis nature of these studies whether they were folded into the current meta-analysis.  In any case, I think that adding two other studies with negative results might have influenced the results (with more or less variants).

It is difficult to determine much from these results.  Like many recent GWAS studies, the authors seem to put a lot of stock in whether the "direction of effect" of the variants remains the same, to the point of claiming this as replication.  I think this is a bit tricky when the variants are considered non-significant in one study, but still accepted as going in the right direction compared to another.  Whatever your interpretation or assessment of the importance of such a finding, I don't think it is fair to call it replication.

Addendum:  In case I didn't make this clear, there are NO genetic variants for depression meeting significance in a study and being independently replicated in another study.  Study design, with overlapping meta-analysis and the use of "confirmation" without doing separate GWAS's make it almost impossible to even try to legitimately replicate, which perhaps is the point.

No comments:

Post a Comment