search

thibautjombart / adegenet

adegenet: a R package for the multivariate analysis of genetic markers
169 stars 64 forks source link

Large dataset syndrome NA values #282

Open dmacguigan opened 4 years ago

dmacguigan commented 4 years ago

Hello,

I am attempting to use the snapclust function in Adegenet v.2.1.3. Unfortunately, I've encountered the following warning message, as discussed by a previous poster (https://github.com/thibautjombart/adegenet/issues/221)

Large dataset syndrome:
 for 41 individuals, differences in log-likelihoods exceed computer precision;
 group membership probabilities are approximated
 (only trust clear-cut values)

snapclust produces NA values for the membership probabilities of those 41 individuals. Oddly, my dataset is not particularly large, especially compared to those mentioned in the previous issue. I have 95 individuals and 3,726 diploid markers.

I have attached my data and code. Please let me know if you have any suggestions for a workaround.

test.zip

dempc01 commented 4 years ago

Hello,

This is an issue I am having as well:

"Large dataset syndrome: for 95 individuals, differences in log-likelihoods exceed computer precision; group membership probabilities are approximated (only trust clear-cut values)"

I am wondering if you have found a solution.

Thanks

zkamvar commented 4 years ago

The "large" in the message is referring to the number of markers, not the number of individuals. I had proposed a potential solution in https://github.com/thibautjombart/adegenet/issues/221#issuecomment-366788840, but neither Thibaut nor I have had time to work on a solution. You might be able to reduce the number of markers you are working with by using PCA to filter the ones that don't contribute much to the first 90% of the variance.

brpark29 commented 4 years ago

Hello! I saw the same error message with one of my rather small datasets (677 samples, ~300 snps). Interestingly, in contrast to the other posters, I did get membership probability estimates for all of my samples. Should I be concerned about receiving this message?