The phylogeny in Figure 2 is based on a consensus tree from 100 bootstrap samples using the "protdist" function and neighboring-joining algorithm in Phylip using the default options.
We used two measures of admixture: the admixture coefficient mY [44] and expected homozygosity FS. mY coefficients and standard deviations were computed as averages of 1,000 random bootstrap samples using the programme ADMIX [44], [45].
Internally validated performance measures will be obtained using 200 bootstrap samples, using the procedure described by Steyerberg.
We therefore approximated the sum and associated 95% confidence band with 1000 bootstrap samples, using the 50th, 2.5th, and 97.5th centiles, respectively.
In addition, we estimate distribution parameters for each of five bootstrap samples, use the estimated parameters to impute values < LOD for the original sample from which we generated the five bootstrap samples, and combine results across the five imputed data sets to obtain MI estimates.
An initial investigation into this was made by repeating simulations for scenario 14 (for which the Branson & Whitehead method previously gave a relatively low coverage of 82.6%) with confidence intervals calculated from 100 bootstrapped samples using the normal approximation method.
However, the bootstrap sampling using 2000 case resampled datasets resulted in a bias corrected confidence interval of 0.558 0.98787, suggesting nevertheless a higher predictive value of these 3 genes than a combination of traditional parameters such as creatinine and INSCO.
To illustrate the utility of adding features not included in our previous models, these r 2 values were compared to those obtained when re-training the multiple linear regression coefficients on each bootstrap sample using only the features of either the context-only or the context+ model, and computing r 2 values on the corresponding test sets.
Confidence intervals were obtained by drawing 10,000 bootstrap samples, using transects as the sampling units for bootstrapping purposes.
QTL mapping was conducted on each bootstrap sample using HSQM, and the position with most significant signal stored for each sample.
