Closed Herest closed 11 months ago
Yeah, you can't have these NaN
values in the correlation matrix.
It usually means you have variants with no variation at all (MAF = 0); you should remove those (cf. similar issues).
Thank you for your response, I filtered the data with plink to keep only those variants with MAF greater than 0.01, but I got the same result when calculating the correlation matrix and in the training of the model. Any other suggestion?
Please have a look at Matrix::which(is.nan(corr0), arr.ind = TRUE)
.
And report the MAF (e.g. with snp_MAF()
for the variant(s) having all NaNs.
Hi again Florian, I fixed the warning about the NAs in the correlation matrix, however, I'm still unable to train the model with auto mode. I keep getting only NAs
Could you say what was the issue, for the record, for people having a similar issue in the future.
For the problem of getting NAs with LDpred2-auto, this has been discussed in other issues here; have you looked at those? Please comment there.
The issue was that there were many SNP with MAF == 0 and missing MAF (i.e. NA). I filtered them using Matrix::which(is.nan(corr0), arr.ind = TRUE)
along with snp_MAF()
.
I also fixed the other issue with the snp_ldpred2_auto
function using the solution in this comment.
Thanks a lot for your help
Hi Florian,
I'm using LDpred2 for training a T2D model using the UK Bio bank data and summary statistics downloaded from Finngen. I followed the steps written in the most recent tutorial from your GitHub and for the QC of the summary statistics I followed the script you have in another repo, however I'm getting all NAs in the final model using the _snp_ldpred2auto.
Do you have idea what am I doing wrong? I'm also getting a warning when calculating the LD correlation matrix
Warning message: NA or NaN values in the resulting correlation matrix.
that was not appearing in other tests that I preformed before doing this training. Thanks @privefl