Journal: BMC Bioinformatics

Article Title: EnzML: multi-label prediction of enzyme classes using InterPro signatures

doi: 10.1186/1471-2105-13-61

Figure Lengend Snippet: Comparison with InterPro2GO2EC and testing on TrEMBL. Left panel: The results of the internal cross-evaluation of the entire SwissProt KEGG and Swiss-Prot datasets are compared with the direct transitive annotation using InterPro2GO and EC2GO lists. The results of training on the SwissProt KEGG dataset and testing on the TrEMBL KEGG dataset are also included. The x axis (accuracy, precision, recall) starts at 50%. Right panel: Comparison of the EC digits in the predicted and actual EC numbers for the TrEMBL KEGG dataset. All predictions all the EC annotations emitted by training on SwissProt KEGG and predicting the unlabelled TrEMBL KEGG (true positives, true negatives, false positives, false negatives). Correct predictions only the predictions corresponding to true, correct annotations existing in TrEMBL KEGG (true positives and true negatives). Wrong predictions false positives and false negatives. The data files used ( SwissProt KEGG and TrEMBL KEGG ) are available as Additional file . The full cross evaluation results are available in Additional file : all_cross_evaluation_results.csv

Article Snippet: As an alternative to machine learning, EC labels could be directly assigned from InterPro domains: the InterPro2GO file associates individual InterPro signatures with GO terms, which in turn are mapped to EC numbers in the EC2GO file.

Techniques: