RomoL2
commented
8 months ago For the first question, that is a great observation- I think the supplemental figure legend may have a typo- the cut off should have been 0.1 for the eQTLs. That cutoff optimized sensitivity/specificity. You can try lower cutoffs which will increase sensitivity but decrease specificity and vice versa. But that cutoff (0.1) was the inflection point to maximize both.
- how would you define or prioritize a 3'UTR variant to be pathogenic based on the annotation output columns? I think that's a bit up to you- we found the strongest predictor of a ClinVar variant being pathogenic was if it is in an eCLIP peak, but there is still signal for other motifs like RBP motifs or microRNAs or even if the variant is predicted by the model to be an eQTL/GWAS variant (see last figure).
- a lot of my variants (all predicted to be 3'UTR by VEP) are lost in the RegVar output. Could you provide some insights into what this might happen? We defined the canonical 3'UTR coordinates using PAS data from Lianoglou et al 2013 (https://pubmed.ncbi.nlm.nih.gov/24145798/). If your variants are outside these coordinates they may be lost. Can you give me an example and I'll verify?
- The output has "tmp_key" column also in there, which I think is why I'm seeing a lot of duplicated rows. Is this supposed to happen? Probably not haha. It looks like a temporary column employed for an intersect that didn't get taken out of the output. I'll work on fixing that and updating the github code.
Lindsay
On Tue, Mar 5, 2024 at 3:24 PM pamelaxu213 @.***> wrote:
Hi Lindsay,
I noticed that in the paper you stated "A variant is predicted to be an eQTL or GWAS hit if its log-odds is greater than 0.01 (eQTL) or 0.0075 (GWAS) ". However, in the CharacterizeVariants.R script line 623 the cutoff seem to be 0.1 for eQTLs. I'm wondering how you decided on the cutoff, and which one I should be using?
A few other questions I have are:
- how would you define or prioritize a 3'UTR variant to be pathogenic based on the annotation output columns?
- a lot of my variants (all predicted to be 3'UTR by VEP) are lost in the RegVar output. Could you provide some insights into what this might happen?
- The output has "tmp_key" column also in there, which I think is why I'm seeing a lot of duplicated rows. Is this supposed to happen?
Thanks for your work and thanks in advance!
— Reply to this email directly, view it on GitHub https://github.com/RomoL2/RegVar/issues/8, or unsubscribe https://github.com/notifications/unsubscribe-auth/AZOARDD5VQXFHS3PK2DWBALYWYSWPAVCNFSM6AAAAABEHZZQT6VHI2DSMVQWIX3LMV43ASLTON2WKOZSGE3TAMBYGYYDQNQ . You are receiving this because you are subscribed to this thread.Message ID: @.***>
pamelaxu213
Hi Lindsay,
I noticed that in the paper you stated "A variant is predicted to be an eQTL or GWAS hit if its log-odds is greater than 0.01 (eQTL) or 0.0075 (GWAS) ". However, in the CharacterizeVariants.R script line 623 the cutoff seem to be 0.1 for eQTLs. I'm wondering how you decided on the cutoff, and which one I should be using?
A few other questions I have are:
Thanks for your work and thanks in advance!