Thrilled to see this out. What started out as a chat several years back with @drfejzo.bsky.social about leveraging publicly available data on hyperemesis gravidarum GWAS turned into a wonderful collaboration with April Shu, @mvaudel.bsky.social, @xwww.bsky.social and many others!
rdcu.be/fdl9k
Some slides from a recent talk on missing heritability.
www.dropbox.com/scl/fi/kvogj...
Gonna be some rad science -- @mollyschumer.bsky.social keynote and a bunch of good talks and posters. And I'll personally pay the registration of anyone who wants to come (it's free).
Differences in evolutionary parameters have important implications for interpreting GWAS results. With the inferred parameters, we reproduced the period when major depression yielded no GWAS discoveries, and we predict that psychiatric disorders will eventually reach higher saturation points.
🧵12/n
This work was done in collaboration with my wonderful mentors @yuvalsim.bsky.social , @jeffspence.github.io , @gs2747.bsky.social , and @jkpritch.bsky.social .
doi.org/10.64898/202...
🧵2/n
Monthly median Received to Accepted time (days) at Nature Genetics
Why do schizophrenia GWAS signals look so flat across the genome?
In our recent preprint, we explored why psychiatric disorders — and, more broadly, brain-related traits involving the central nervous system — appear to have unusual genetic architectures.
🧵1/n
We invite you to join our seminar at UCSF Mission Bay campus with Hakhamanesh Mostafavi from NYU Grossman School of Medicine
humangenetics.ucsf.edu/ihg-seminar-...
It's out! I hope this work encourages folks to move beyond a standard "one variant, one gene" QTL paradigm and consider proxitropic variant effects. Big thanks to the reviewers and editors at @ajhgnews.bsky.social for their help! @sbmontgom.bsky.social www.sciencedirect.com/science/arti...
Excited to share our new preprint from the @arbelharpak.bsky.social Lab!
How do recruitment into genetic studies and study characteristics impact what we infer about the genetic bases of traits, and what are the consequences? (1/21)
www.biorxiv.org/content/10.6...
If you like larger sample sizes, then do check out our reprocessed and fine mapped cis-eQTLs and cis-sQTLs (leafCutter and MAJIQ!) from the INTERVAL cohort (whole blood, n up to 4,729)!
zenodo.org/records/1795...
These will be on the eQTL Catalogue FTP soon as well.
cc @yosephbarash.bsky.social
We used a similar design in this paper:
www.sciencedirect.com/science/arti...
Found that in cis-MR with multiple colocalising instruments (i.e. multiple indepdent signals colocalise between eQTL and pQTL traits) sign concordance was around 95%:
Thank you for pointing these evaluations out!
Your numbers after stringent colocalization seem quite consistent with theirs (tagging @nikhilmilind.dev @jkpritch.bsky.social) . Colocalization is clearly needed, but it kills recall and that's a difficult thing to get around with standard sample sizes...
Clever use of proteomic data to stress-test TWAS and QTL colocalization methods, revealing a high false sign rate. This hypothesis about high-LD and cross-tissue confounding is particularly interesting:
Happy to share our new preprint from @sashagusevposts.bsky.social and @nmancuso.bsky.social labs! We introduce Mr. PEG, a framework integrating perturbational screens, eQTL, and GWAS data to identify mediating genes for complex traits. (1/n) www.medrxiv.org/content/10.6...
New method from our group for identifying disease-mediating genes using perturb-seq, eQTL, and GWAS data. Check out the thread:
New preprint alert: we use sign errors as a test of how well TWAS works.
Very worryingly we find that TWAS gets the sign wrong around 1/3 of the time (compared to 50% for pure guessing). You can read more about our analysis here, and what we think is going on 👇
Hope you enjoy reading our pre-print, and we are happy to take any feedback!
doi.org/10.64898/202...
19/19
It was a pleasure to work with co-first-author @peter-gerlach.bsky.social, who led the proteomics analysis, along with our mentors @jeffspence.github.io and @jkpritch.bsky.social!
18/n
See more about this here:
bsky.app/profile/jeff...
17/n
Although burden tests are a gold standard, various factors such as specificity and constraint reduce our power from rare-variant burden tests. Common variants, which are context-specific, provide critical information about the direction-of-effect.
16/n
The active development of TWAS-type methods continues to be critical for summarizing the direction-of-effect from common variants.
15/n
These results suggest that TWAS often nominates the correct gene for the wrong reasons. When GWAS signal is present, a nearby gene is likely to be involved, but the wrong tissue may have a confident TWAS association with a random sign due to strong linkage.
14/n
We dug into the association of GWAS variants at the GCK locus with blood glucose levels. There was little eQTL signal in the relevant pancreatic tissue, but strong associations in the incorrect direction with evidence of colocalization in tibial nerve and thyroid tissue.
13/n
The false sign rate drops substantially if we analyze the protein data using the relevant tissue. This suggests that tightly-linked eQTL in irrelevant tissues may show strong TWAS and colocalization signals, even though they are not mechanistically related to the GWAS variant.
12/n
Similar to the proteomics analysis, we found that around 33% of genes had an inconsistent direction-of-effect.
11/n
We conducted the same analysis using complex traits. For each gene, we expected the TWAS direction-of-effect to match the effect inferred from loss-of-function variants, which are often used as a gold standard for estimating gene effect direction.
10/n
TWAS methods are known to have high false-positive rates, and evidence of colocalization is often used as a filtering strategy. However, even with extremely high evidence of colocalization, the false sign rate remained at around 10%, and resulted in a large drop in recall.
9/n
This is a much higher false sign rate than rare-variant burden tests using LoF variants or duplications, which tend to correctly assign the direction-of-effect for the same proteomics data.
8/n