Chromosome specific reference panel and some questions

[huilisabrina]

Hi @zhenin ,

Thanks again for making this software available. I was wondering if it'd be possible to add an option to the package for chromosome-specific heritability and rg estimation? This will make simulation or method comparisons easier. I know you're planning to release scripts for reference panel generation. I think that will solve the issue ultimately as well, but just wondering if it'd be possible to enable chromosome-wide estimation (based on your existing ref panel) in the interim?

Another feedback is that I noticed in some of the log files that the report of SNP proportions looks funny. For example, there were some cases where the percent of GWAS SNPs in the reference exceeds 100%:

Analysis starts on Thu Feb 11 21:59:27 2021 
307871 out of 307519 (100.11%) SNPs in reference panel are available in GWAS 1.  
251057 out of 307519 (81.64%) SNPs in reference panel are available in GWAS 2.  
Warning: More than 1% SNPs in reference panel are missed in GWAS 2. This may generate bias in estimation. Please make sure that you are using correct reference panel.  

This makes me wonder if the proportion was calculated based on the overlapped SNPs (between the study sample and the ref panel) or if there's any issues with the merging codes.

Finally, I was also getting some heritability estimates that were 1 with s.e. 0. (for reference I used the same set of association results and got 0.2146 (0.0681) from LDSC). I'm guessing the point estimate of 1 was not the actual HDL estimate itself but may due to numerical under/overflow during the calculation process? I'm not sure, but would it be possible to catch that and break the calculation by giving a warning?

Thanks in advance for your help!
Hui

[zhenin]

Hi @huilisabrina ,

Thanks a lot for your feedback!

1. Chromosome-wide HDL could be interesting. I will add it in the next update.
2. I realize there is a bug in the merging codes when there are duplicated or multiallelic SNPs. This will be fixed in the next update.
3. This is a bit weird and worth investigating. Which reference panel are you using? Could you paste the full report here?

Best,
Zheng

[huilisabrina]

@zhenin Thanks for your responses!

Certainly, I tried using all three different reference panels and still got the 1(0) estimates for these two summary statistics: Coronary artery disease (Nelson et al. 2017) and Type 2 Diabetes (Wood et al. 2016). I can email you the formatted sumstats (input to HDL) directly if that's easier to look into.

Thanks,
Hui

[zhenin]

@zhenin Thanks for your responses!

Certainly, I tried using all three different reference panels and still got the 1(0) estimates for these two summary statistics: Coronary artery disease (Nelson et al. 2017) and Type 2 Diabetes (Wood et al. 2016). I can email you the formatted sumstats (input to HDL) directly if that's easier to look into.

Thanks,
Hui

That would be great! My email is zheng.ning@ki.se.

Best,
Zheng