Skip to main content
. 2019 Oct 11;36(5):1429–1438. doi: 10.1093/bioinformatics/btz762

Table 4.

Variable importance percentages grouped by feature classes for the BCrystal model highlighting the feature contributions in order as depicted in Figure 2

Feature class Total features Total features (>0) Total imp (%)
Log(L) 1 1 0.052
Log(MW) 1 1 0.032
Turn Freq 1 0 0.000
Gravy index 1 1 0.017
Aliphatic index 1 0 0.000
Total charge 1 0 0.000
Mono Freq SS3 3 1 0.068
Di Freq SS3 9 0 0.000
Tri Freq SS3 27 3 0.063
Mono Freq SS8 8 4 0.664
Di Freq SS8 64 8 1.051
Tri Freq SS8 512 18 1.001
FER at RSA cutoffs 20 10 81.08
FER at RSA cutoff × HP 20 12 11.53
Disorder features 25 5 1.906
PBS in disorder features 25 2 0.099
Mono Freq AA 20 7 0.479
Di Freq AA 400 31 1.018
Tri Freq AA 8000 23 0.937
All features 9139 127 100.00

Note: Here, total features (>0) represents the total number of variables from a feature class with non-zero variable importance scores.