Table 1.
Illustration of the three different encoding schemes for SNP data.
Table 2.
Algorithm implementations.
Table 3.
Parameter optimization values.
Table 4.
Average number of SNPs reaching the specified p-value threshold per data set.
Table 5.
Average ranks and p-values of the Friedman test for the three encoding schemes.
Table 6.
Rank differences and p-values for pair-wise comparison of encodings.
Fig 1.
Comparison of encodings per classifier.
The three encodings compared by their rank distance over all data sets and classifiers (a) and grouped by classifier. A connecting line between encodings means that the null hypothesis of them being significantly different could not be rejected. Only algorithms for which the Friedman test rejected the null hypothesis are shown. (α = 0.001.)
Fig 2.
Comparison of encodings per disease data set.
The three encodings compared by their rank distance over all data sets and classifiers (a) and grouped by disease data set. A connecting line between encodings means that the null hypothesis of them being significantly different could not be rejected. Only data sets for which the Friedman test rejected the null hypothesis are shown. (α = 0.001.)
Table 7.
Maximum and average AUCs for different encodings grouped by data set.
Table 8.
Average ranks of the seven classification algorithms.
Table 9.
Rank differences and p-values for pair-wise comparison of classification algorithms.
Fig 3.
Comparison of classification algorithms.
The seven classification algorithms compared by their rank distance over all disease data sets using the additive encoding. A connecting line between encodings means that the null hypothesis of them being significantly different could not be rejected (with α = 0.001).
Table 10.
Average AUC for each data set and algorithm over all p-value thresholds.