EFICAz2: enzyme function inference by a combined approach enhanced by machine learning

Abstract

Background: We previously developed EFICAz, an enzyme function inference approach that combines predictions from non-completely overlapping component methods. Two of the four components in the original EFICAz are based on the detection of functionally discriminating residues (FDRs). FDRs distinguish between member of an enzyme family that are homofunctional (classified under the EC number of interest) or heterofunctional (annotated with another EC number or lacking enzymatic activity). Each of the two FDR-based components is associated to one of two specific kinds of enzyme families. EFICAz exhibits high precision performance, except when the maximal test to training sequence identity (MTTSI) is lower than 30%. To improve EFICAz's performance in this regime, we: i) increased the number of predictive components and ii) took advantage of consensual information from the different components to make the final EC number assignment.

Results: We have developed two new EFICAz components, analogs to the two FDR-based components, where the discrimination between homo and heterofunctional members is based on the evaluation, via Support Vector Machine models, of all the aligned positions between the query sequence and the multiple sequence alignments associated to the enzyme families. Benchmark results indicate that: i) the new SVM-based components outperform their FDR-based counterparts, and ii) both SVM-based and FDR-based components generate unique predictions. We developed classification tree models to optimally combine the results from the six EFICAz components into a final EC number prediction. The new implementation of our approach, EFICAz2, exhibits a highly improved prediction precision at MTTSI < 30% compared to the original EFICAz, with only a slight decrease in prediction recall. A comparative analysis of enzyme function annotation of the human proteome by EFICAz2 and KEGG shows that: i) when both sources make EC number assignments for the same protein sequence, the assignments tend to be consistent and ii) EFICAz2 generates considerably more unique assignments than KEGG.

Conclusion: Performance benchmarks and the comparison with KEGG demonstrate that EFICAz2 is a powerful and precise tool for enzyme function annotation, with multiple applications in genome analysis and metabolic pathway reconstruction. The EFICAz2 web service is available at: http://cssb.biology.gatech.edu/skolnick/webservice/EFICAz2/index.html.

Figure 1

Prediction performance of the FDR-based and SVM-based approaches applied to Multiple Pfam enzyme families. For three-field (A, B) or four-field EC number classifiers (C, D), the average recall (A, C) and average precision (B, D) of the FDR-based (blue columns) and SVM-based (red columns) approaches is plotted at different intervals of maximal test to training sequence identity (MTTSI). The average of each performance indicator is done over all the EC numbers defined in the specified MTTSI interval (numbers at the bottom of each column). Details about the benchmark can be found in "Benchmarking of EFICAz² version 10", in the Methods section. Statistically significant differences in performance are indicated by black lines under the corresponding columns (see "Statistical analyses", in the Methods section). Values on top of each column represent average +/- standard deviation.

Figure 2

Prediction performance of the FDR-based and SVM-based approaches applied to CHIEFc enzyme families. For three-field (A, B) or four-field EC number classifiers (C, D), the average recall (A, C) and average precision (B, D) of the FDR-based (blue columns) and SVM-based (red columns) approaches is plotted at different intervals of maximal test to training sequence identity (MTTSI). The average of each performance indicator is done over all the EC numbers defined in the specified MTTSI interval (numbers at the bottom of each column). Details about the benchmark can be found in "Benchmarking of EFICAz² version 10", in the Methods section. Statistically significant differences in performance are indicated by black lines under the corresponding columns (see "Statistical analyses", in the Methods section). Values on top of each column represent average +/- standard deviation.

Figure 3

Prediction overlap of FDR-based and SVM-based methods. The fractions of test sequences (corresponding to the benchmark described in "Benchmarking of EFICAz² version 10", in the Methods section) correctly predicted by three or four-field EC number classifiers applied to Multiple Pfam or CHIEFc enzyme families are represented. For combination of enzyme family and level of description of the classifiers, we show the fraction corresponding to unique predictions made by the FDR-based (blue) or SVM-based method (green), and the fraction corresponding to predictions made by both (orange) or none of the methods (yellow).

Figure 4

Prediction performance of different EFICAz implementations. For three-field (A, B) or four-field EC number classifiers (C, D), the average recall (A, C) and average precision (B, D) of the original EFICAz (green columns), EFICAz plus the new SVM-based components (blue columns) and EFICAz² (red columns) is plotted at different intervals of maximal test to training sequence identity (MTTSI). The average of each performance indicator is done over all the EC numbers defined in the specified MTTSI interval (numbers at the bottom of each column). Details about the benchmark can be found in "Benchmarking of EFICAz² version 10", in the Methods section. Statistically significant differences in performance are indicated by black lines under the corresponding columns (see "Statistical analyses", in the Methods section). Values on top of each column represent average +/- standard deviation.

Figure 5

Predictive models for EFICAz² based on classification trees. Classification trees corresponding to three-field (A, B) and four-field EC numbers (C, D) to integrate predictions from each of the six EFICAz² components for protein sequences that exhibit MTTSI < 30% (A, C) or MTTSI ≥ 30% (B, D). CH_FDR = CHIEFc family based FDR recognition; PF_FDR = Multiple Pfam family based FDR recognition; CH_SIT = CHIEFc family specific SIT evaluation; Prst = High specificity multiple PROSITE pattern recognition; CH_svm = CHIEFc family based SVM evaluation; PF_svm = Multiple Pfam family based SVM evaluation.

Figure 6

Distribution of the number of test sequences per enzyme type. Distribution of 9,397 test enzyme sequences into 145 types of three-field EC numbers (green columns) and 6,996 test enzyme sequences into 614 types of four-field EC numbers (red columns).