Enzyme function prediction with interpretable models

Application of a hierarchical enzyme classification method reveals the role of gut microbiome in human metabolism

Background

Enzymes are known as the molecular machines that drive the metabolism of an organism; hence identification of the full enzyme complement of an organism is essential to build the metabolic blueprint of that species as well as to understand the interplay of multiple species in an ecosystem. Experimental characterization of the enzymatic reactions of all enzymes in a genome is a tedious and expensive task. The problem is more pronounced in the metagenomic samples where even the species are not adequately cultured or characterized. Enzymes encoded by the gut microbiota play an essential role in the host metabolism; thus, warranting the need to accurately identify and annotate the full enzyme complements of species in the genomic and metagenomic projects. To fulfill this need, we develop and apply a method called ECemble, an ensemble approach to identify enzymes and enzyme classes and study the human gut metabolic pathways.

Results

ECemble method uses an ensemble of machine-learning methods to accurately model and predict enzymes from protein sequences and also identifies the enzyme classes and subclasses at the finest resolution. A tenfold cross-validation result shows accuracy between 97 and 99% at different levels in the hierarchy of enzyme classification, which is superior to comparable methods. We applied ECemble to predict the entire complements of enzymes from ten sequenced proteomes including the human proteome. We also applied this method to predict enzymes encoded by the human gut microbiome from gut metagenomic samples, and to study the role played by the microbe-derived enzymes in the human metabolism. After mapping the known and predicted enzymes to canonical human pathways, we identified 48 pathways that have at least one bacteria-encoded enzyme, which demonstrates the complementary role of gut microbiome in human gut metabolism. These pathways are primarily involved in metabolizing dietary nutrients such as carbohydrates, amino acids, lipids, cofactors and vitamins.

Conclusions

The ECemble method is able to hierarchically assign high quality enzyme annotations to genomic and metagenomic data. This study demonstrated the real application of ECemble to understand the indispensable role played by microbe-encoded enzymes in the healthy functioning of human metabolic systems.

Prediction of detailed enzyme functions and identification of specificity determining residues by random forests

Determining enzyme functions is essential for a thorough understanding of cellular processes. Although many prediction methods have been developed, it remains a significant challenge to predict enzyme functions at the fourth-digit level of the Enzyme Commission numbers. Functional specificity of enzymes often changes drastically by mutations of a small number of residues and therefore, information about these critical residues can potentially help discriminate detailed functions. However, because these residues must be identified by mutagenesis experiments, the available information is limited, and the lack of experimentally verified specificity determining residues (SDRs) has hindered the development of detailed function prediction methods and computational identification of SDRs. Here we present a novel method for predicting enzyme functions by random forests, EFPrf, along with a set of putative SDRs, the random forests derived SDRs (rf-SDRs). EFPrf consists of a set of binary predictors for enzymes in each CATH superfamily and the rf-SDRs are the residue positions corresponding to the most highly contributing attributes obtained from each predictor. EFPrf showed a precision of 0.98 and a recall of 0.89 in a cross-validated benchmark assessment. The rf-SDRs included many residues, whose importance for specificity had been validated experimentally. The analysis of the rf-SDRs revealed both a general tendency that functionally diverged superfamilies tend to include more active site residues in their rf-SDRs than in less diverged superfamilies, and superfamily-specific conservation patterns of each functional residue. EFPrf and the rf-SDRs will be an effective tool for annotating enzyme functions and for understanding how enzyme functions have diverged within each superfamily.

SHARP: genome-scale identification of gene-protein-reaction associations in cyanobacteria

Genome scale metabolic model provides an overview of an organism's metabolic capability. These genome-specific metabolic reconstructions are based on identification of gene to protein to reaction (GPR) associations and, in turn, on homology with annotated genes from other organisms. Cyanobacteria are photosynthetic prokaryotes which have diverged appreciably from their nonphotosynthetic counterparts. They also show significant evolutionary divergence from plants, which are well studied for their photosynthetic apparatus. We argue that context-specific sequence and domain similarity can add to the repertoire of the GPR associations and significantly expand our view of the metabolic capability of cyanobacteria. We took an approach that combines the results of context-specific sequence-to-sequence similarity search with those of sequence-to-profile searches. We employ PSI-BLAST for the former, and CDD, Pfam, and COG for the latter. An optimization algorithm was devised to arrive at a weighting scheme to combine the different evidences with KEGG-annotated GPRs as training data. We present the algorithm in the form of software "Systematic, Homology-based Automated Re-annotation for Prokaryotes (SHARP)." We predicted 3,781 new GPR associations for the 10 prokaryotes considered of which eight are cyanobacteria species. These new GPR associations fall in several metabolic pathways and were used to annotate 7,718 gaps in the metabolic network. These new annotations led to discovery of several pathways that may be active and thereby providing new directions for metabolic engineering of these species for production of useful products. Metabolic model developed on such a reconstructed network is likely to give better phenotypic predictions.

A top-down approach to classify enzyme functional classes and sub-classes using random forest

Advancements in sequencing technologies have witnessed an exponential rise in the number of newly found enzymes. Enzymes are proteins that catalyze bio-chemical reactions and play an important role in metabolic pathways. Commonly, function of such enzymes is determined by experiments that can be time consuming and costly. Hence, a need for a computing method is felt that can distinguish protein enzyme sequences from those of non-enzymes and reliably predict the function of the former. To address this problem, approaches that cluster enzymes based on their sequence and structural similarity have been presented. But, these approaches are known to fail for proteins that perform the same function and are dissimilar in their sequence and structure. In this article, we present a supervised machine learning model to predict the function class and sub-class of enzymes based on a set of 73 sequence-derived features. The functional classes are as defined by International Union of Biochemistry and Molecular Biology. Using an efficient data mining algorithm called random forest, we construct a top-down three layer model where the top layer classifies a query protein sequence as an enzyme or non-enzyme, the second layer predicts the main function class and bottom layer further predicts the sub-function class. The model reported overall classification accuracy of 94.87% for the first level, 87.7% for the second, and 84.25% for the bottom level. Our results compare very well with existing methods, and in many cases report better performance. Using feature selection methods, we have shown the biological relevance of a few of the top rank attributes.

Computational Approaches for Automated Classification of Enzyme Sequences

Determining the functional role(s) of enzymes is very important to build the metabolic blueprint of an organism and to identify the potential roles enzymes may play in metabolic and disease pathways. With exponential growth in gene and protein sequence data, it is not feasible to experimentally characterize the function(s) of all enzymes. Alternatively, computational methods can be used to annotate the enormous amount of unannotated enzyme sequences. For function prediction and classification of enzymes, features based on amino acid composition, sequence and structural properties, domain composition and specific peptide information have been widely used by different computational approaches. Each feature space has its own merits and limitations on the overall prediction accuracy. Prediction accuracy improves when machine-learning methods are used to classify enzymes. Given the incomplete and unbalanced nature of annotations in biological databases, ensemble methods or methods that bank on a combination of orthogonal feature are more desirable for achieving higher accuracy and coverage in enzyme classification. In this review article, we systematically describe all the features and methods used thus far for enzyme class prediction. To the authors' knowledge, this review represents the most exhaustive description of methods used for computational prediction of enzyme classes.

Detecting subtle functional differences in ketopantoate reductase and related enzymes using a rule-based approach with sequence-structure homology recognition scores

Ketopatoate reductase (KPR) is the second enzyme in the pantothenate (vitamin B(5)) biosynthesis pathway, an essential metabolic pathway identified as a potential target for new antimicrobials. The sequence similarity among putative KPRs is limited and KPR itself belongs to a large superfamily of 6-phosphogluconate dehydrogenases. Therefore, it is necessary to discriminate between true and other enzymes. In this paper, we describe a systematic analysis of putative KPRs in the context of this superfamily. Detailed structural analysis allowed us to define key residues for KPR activity and we classified eight structural genomics structures of the KPR family into four functional subclasses. We proposed a semi-automatic protocol, using sequence-structure homology recognition scores, for assigning KPR and related proteins to these subclasses and applied it to a representative set of 103 completely sequenced bacterial genomes. A similar approach can be applied to other enzyme families, which would aid the correct identification of drug targets and help design novel specific inhibitors.

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

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.