Erratum to: genetic algorithm learning as a robust approach to RNA editing site site prediction

BMC Bioinformatics

Table 1 Overall performance of REGAL on A. thaliana.

	Known Edited Sites Total: 17 – 26	Known Unedited Sites Total: 18 – 28
Predicted Edited Site	True positive 19.4 (± 3.4)	False positive 3.3 (± 1.2)	Sensitivity: 0.91 (± 0.06) Specificity: 0.85 (± 0.06)
Predicted Unedited Site	False negative 2.0 (± 1.1)	True negative 19.7 (± 3.8)	PPV: 0.86 (± 0.05) Accuracy: 0.88 (± 0.05)

We tested the performance of REGAL on known edited and unedited sites from three mitochondrial genomes. The results from A. thaliana were obtained after 10 iterations of cross-validation using on average 33 edited and 33 unedited sites per testing data set (see Implementation for details of 10-fold cross-validation). The overall accuracy in this genome was 81%, with sensitivity of 81% and specificity of 80%. Within the 90% credible intervals, on average 22 edited sites and 23 unedited sites were assessed. We report the range of values as obtained from the cross-validation. Since the proportion of true positives to true negatives varied slightly in each test data set, we report both specificity and positive predictive value (PPV).

ISSN: 1471-2105