Strange (and unrealistic) fastSTRUCTURE assignment results?

[bbarker505]

Hi all,

I'm a little perplexed by my results from fastSTRUCTURE. I ran STRUCTURE 2.3.4 and fastSTRUCTURE on the same SNP dataset (~10K SNPs). The assignment probabilities for each sample according to STRUCTURE make complete sense, but those from fastSTRUCTURE do not. For example, some of the samples belong to a well-differentiated group that always have an assignment probability of >0.96 of belonging to their own group according to STRUCTURE, but according to fastSTRUCTURE some of those samples are very admixed. I applied a model of correlated allele frequencies and admixture in STRUCTURE and did not use a locality ID prior. I left everything in fastSTRUCTURE as default.

I can only think of a couple of possibilities as to why I might be getting weird results, and was hoping that readers might have some more ideas. 

1) missing data? Some of the samples do have quite a bit of missing data - e.g. as high as about 30% for some loci, but most are <10%

2) using the simple algorithm when perhaps I should use the logistic algorithm?

I was hoping to use this program because STRUCTURE is so slow and it wont' run on newer Ubuntu operating systems (to my knowledge), so thanks for any help!

Brit

[bbarker505]

I am answering my own question. It turns out that all missing data must be coded as -9 in FastStructure. I had been coding missing data as 0.

[LR]

Thanks bbarker505!
I am running FasStructure for the first time and my results are completely different from PCA and preliminary Structure results, this could be the reason!