Table 1.
Results on Arabidopsis thaliana
| method | train | test | nucAcc | exonF | geneSn |
| GRAPE | CV | CV | 95 ± 1% | 82 ± 2% | 49 ± 3% |
| GRAPE | CV | H | 93 ± 1% | 80 ± 2% | 44 ± 3% |
| GRAPE | T | T | 95% | 86% | 57% |
| GRAPE | T | H | 94% | 81% | 48% |
| MLE | CV | CV | 90 ± 1% | 72 ± 2% | 33 ± 4% |
| MLE | T | T | 91% | 75% | 36% |
| MLE | T | H | 90% | 71% | 33% |
GRAPE = GRadient Ascent Parameter Estimation, MLE = Maximum Likelihood Estimation only. CV=cross validation, T = training set, H = 1000-gene hold-out ("test") set. CV in the train column means training on 800 genes from T. CV in test column means testing on 200 genes from T. In rows with a CV in either column, numbers are averages from 5 runs. nucAcc = nucleotide accuracy, exonF = exon F score, geneSn = gene sensitivity. F = 2SnSp/(Sn+Sp) for Sn = sensitivity and Sp = specificity. CV averages are reported ± SD.