Table 5.
Effect of parameter set indexing strategy on PFCWLLKR performance using non-TIS-containing data
| k | Indexing strategy | TN | FP | Sn | |
| 3 | modulating | edit | 5,074 | 11,047 | 0.3147 |
| PWM | 5,134 | 10,987 | 0.3185 | ||
| WAM | 5,069 | 11,052 | 0.3144 | ||
| static | edit | 6,170 | 9,951 | 0.3827 | |
| PWM | 6,279 | 9,842 | 0.3895 | ||
| WAM | 6,119 | 10,002 | 0.3796 | ||
| 5 | modulating | edit | 4,537 | 11,584 | 0.2814 |
| PWM | 4,484 | 11,637 | 0.2781 | ||
| WAM | 4,262 | 11,859 | 0.2644 | ||
| static | edit | 5,993 | 10,128 | 0.3718 | |
| PWM | 6,065 | 10,056 | 0.3762 | ||
| WAM | 5,679 | 10,442 | 0.3523 | ||
| 10 | modulating | edit | 4,190 | 11,931 | 0.2599 |
| PWM | 3,708 | 12,413 | 0.2300 | ||
| WAM | 3,533 | 12,588 | 0.2192 | ||
| static | edit | 6,345 | 9,776 | 0.3936 | |
| PWM | 5,537 | 10,584 | 0.3435 | ||
| WAM | 5,199 | 10,922 | 0.3225 | ||
16,121 non-TIS-containing instances were used in five-fold cross-validation experiments, in which parameter sets were selected for putative TIS evaluation according to best cluster fit established by either the Hamming distance relative to cached medoids (edit), position weight matrix scores (PWM), or weight array matrix scores (WAM). Parameter indexing was tested under both modulating (cluster assignment for each site separately) and static (cluster assignment based on the leftmost ATG) approaches. k denotes the number of clusters considered. TN represents the number of instances for which the method (correctly) refused to predict a TIS, and FP the number for which some prediction was made, though always incorrect (see Figure 2). .