Because it is a full Bayesian model, the BBC can handle missing data extremely easily, and can also incorporate likelihood-based criterion, such as AIC, BIC, maximum likelihood, Bayes factors, etc., for model evaluations and comparisons.
Overall levels of missing data were extremely low for both tools across the sample of 285 individuals.
ASTRAL provides one such advance, but new and more accurate coalescent-based methods are needed to enable these analyses, especially for genome-scale datasets where missing data and extremely low phylogenetic signal in individual genes may be a substantial problem.
Missing data levels were extremely low, typically less than 5% for most fields.
Because of the extremely low rate of missing data to start with, only 7 cases and 89 controls were removed by this process.
Although genotyping microarrays produce some missing data, GBS produces an extremely sparse genotype matrix with mostly missing data: often hundreds of thousands of SNPs are discovered, but only a small fraction of the SNPs (e.g., <10%) pass missing data thresholds (Gardner et al. 2014).
Missing data were extremely rare (four covariates had missing data for one or more case(s), with a maximum of 0.02%), but if present they were coded as dummy variables in the analysis.
4 The overall 1-year mortality rate associated with hip fracture is reported to be between 14% to 47%. 5, 6 Still, there is missing data on functional outcomes and mortality rates in extremely elderly patients.
However, it is extremely common in the pastoralist literature to find missing data points in some analyses due to migration, loss of follow-up, and subject refusal to participate, among other reasons.
Thus, the proportion of missing data in the DPDB (and in turn in the DNPR) is extremely low.
Our missing data, noise, and inbreeding experiments suggest that PSIKO as well as ADMIXTURE and sNMF handle these types of data extremely well.
