LDSC on build38 data

[mvab]

Hi,

Has anyone successfully applied LDSC genetic correlation to build38 GWAS data? (without doing a liftover to build37)

It seems that the required ref data is available here, but currently not sure which specific version. It probably also needs some pre-processing to match the ref files used for build37.

Is anyone aware of any tutorials/ github repos where people have done it?

[aksarkar]

At Vesalius Therapeutics, we have re-built the ldsc baseline and routinely analyze build 38 data.

Unfortunately, we are not able to provide the reference data or code to generate these at this time.

I can outline that you will need to generate annotations, partitioned LD scores (as in https://github.com/bulik/ldsc/wiki/LD-Score-Estimation-Tutorial#partitioned-ld-scores), weights (as in https://github.com/bulik/ldsc/wiki/LD-Score-Estimation-Tutorial#--w-ld), and frequency files anew from (1) a genotype matrix for reference individuals, (2) the baseline BED files https://zenodo.org/records/7768714/files/baselineLD_v2.2_bedfiles.tgz?download=1 (lifted), and (3) the list of well-imputed variants https://zenodo.org/records/7768714/files/1000G_Phase3_EAS_weights_hm3_no_MHC.tgz?download=1 (cross-referenced against dbSNP, then lifted).

[mvab]

Hi @aksarkar,

Thanks for your reply!

Yes I could tell it's not as straightforward as finding the correct files that could be used right away. Thanks for sharing some hints on the processing required.