Enzyme Commission Classification Research Articles

IntEnzyDB: an Integrated Structure-Kinetics Enzymology Database.

Data-driven modeling has emerged as a new paradigm for biocatalyst design and discovery. Biocatalytic databases that integrate enzyme structure and function data are in urgent need. Here we describe IntEnzyDB as an integrated structure-kinetics database for facile statistical modeling and machine learning. IntEnzyDB employs a relational database architecture with a flattened data structure, which allows rapid data operation. This architecture also makes it easy for IntEnzyDB to incorporate more types of enzyme function data. IntEnzyDB contains enzyme kinetics and structure data from six enzyme commission classes. Using 1050 enzyme structure-kinetics pairs, we investigated the efficiency-perturbing propensities of mutations that are close or distal to the active site. The statistical results show that efficiency-enhancing mutations are globally encoded and that deleterious mutations are much more likely to occur in close mutations than in distal mutations. Finally, we describe a web interface that allows public users to access enzymology data stored in IntEnzyDB. IntEnzyDB will provide a computational facility for data-driven modeling in biocatalysis and molecular evolution.

Effects of sequence features on machine-learned enzyme classification fidelity

Assigning enzyme commission (EC) numbers using sequence information alone has been the subject of recent classification algorithms where statistics, homology and machine-learning based methods are used. This work benchmarks performance of a few of these algorithms as a function of sequence features such as chain length and amino acid composition (AAC). This enables determination of optimal classification windows for de novo sequence generation and enzyme design. Parallelization and visualization workflows are developed to observe the performance of the classifier over changing enzyme length, main EC class and AAC. We applied these workflows to the entire SwissProt database to date (n = 565928) using two, locally installable classifiers, ECpred and DeepEC, and collecting results from two other webserver-based tools, Deepre and BENZ-ws. All the classifiers exhibit peak performance in the range of 300–450 amino acids in length. Classifiers were most accurate at predicting translocases (EC-6) and were least accurate in determining hydrolases (EC-3) and oxidoreductases (EC-1). We also identified AAC ranges that are most common in the annotated enzymes and found that all classifiers work best in this common range. Among the four classifiers, ECpred showed the best consistency in changing feature space. These workflows can be used to benchmark new algorithms as they are developed and find optimum design spaces for the generation of new, synthetic enzymes.

Electrostatic fingerprints of catalytically active amino acids in enzymes.

The computed electrostatic and proton transfer properties are studied for 20 enzymes that represent all six major enzyme commission classes and a variety of different folds. The properties of aspartate, glutamate, and lysine residues that have been previously experimentally determined to be catalytically active are reported. The catalytic aspartate and glutamate residues studied here are strongly coupled to at least one other aspartate or glutamate residue and often to multiple other carboxylate residues with intrinsic pKa differences less than 1 pH unit. Sometimes these catalytic acidic residues are also coupled to a histidine residue, such that the intrinsic pKa of the acidic residue is higher than that of the histidine. All catalytic lysine residues studied here are strongly coupled to tyrosine or cysteine residues, wherein the intrinsic pKa of the anion-forming residue is higher than that of the lysine. Some catalytic lysines are also coupled to other lysines with intrinsic pKa differences within 1 pH unit. Some evidence of the possible types of interactions that facilitate nucleophilicity is discussed. The interactions reported here provide important clues about how side chain functional groups that are weak Brønsted acids or bases for the free amino acid in solution can achieve catalytic potency and become strong acids, bases or nucleophiles in the enzymatic environment.

Biocatalysed synthesis planning using data-driven learning

Enzyme catalysts are an integral part of green chemistry strategies towards a more sustainable and resource-efficient chemical synthesis. However, the use of biocatalysed reactions in retrosynthetic planning clashes with the difficulties in predicting the enzymatic activity on unreported substrates and enzyme-specific stereo- and regioselectivity. As of now, only rule-based systems support retrosynthetic planning using biocatalysis, while initial data-driven approaches are limited to forward predictions. Here, we extend the data-driven forward reaction as well as retrosynthetic pathway prediction models based on the Molecular Transformer architecture to biocatalysis. The enzymatic knowledge is learned from an extensive data set of publicly available biochemical reactions with the aid of a new class token scheme based on the enzyme commission classification number, which captures catalysis patterns among different enzymes belonging to the same hierarchy. The forward reaction prediction model (top-1 accuracy of 49.6%), the retrosynthetic pathway (top-1 single-step round-trip accuracy of 39.6%) and the curated data set are made publicly available to facilitate the adoption of enzymatic catalysis in the design of greener chemistry processes.

Experimental phasing with vanadium and application to nucleotide-binding membrane proteins.

The structure determination of soluble and membrane proteins can be hindered by the crystallographic phase problem, especially in the absence of a suitable homologous structure. Experimental phasing is the method of choice for novel structures; however, it often requires heavy-atom derivatization, which can bedifficult and time-consuming. Here, a novel and rapid method to obtain experimental phases for protein structure determination by vanadium phasing is reported. Vanadate is a transition-state mimic of phosphoryl-transfer reactions and it has the advantage of binding specifically to the active site of numerous enzymes catalyzing this reaction. The applicability of vanadium phasing has been validated by determining the structures of three different protein-vanadium complexes, two of which are integral membrane proteins: the rabbit sarcoplasmic reticulum Ca2+-ATPase, the antibacterial peptide ATP-binding cassette transporter McjD from Escherichia coli and the soluble enzyme RNAse A from Bos taurus. Vanadium phasing was successful even at low resolution and despite severe anisotropy in the data. This method is principally applicable to a large number of proteins, representing six of the seven Enzyme Commission classes. It relies exclusively on the specific chemistry of the protein and it does not require any modifications, making it a very powerful addition to the phasing toolkit. In addition to the phasing power of this technique, the protein-vanadium complexes also provide detailed insights into the reaction mechanisms of the studied proteins.

Functional analysis of Rossmann-like domains reveals convergent evolution of topology and reaction pathways.

Rossmann folds are ancient, frequently diverged domains found in many biological reaction pathways where they have adapted for different functions. Consequently, discernment and classification of their homologous relations and function can be complicated. We define a minimal Rossmann-like structure motif (RLM) that corresponds for the common core of known Rossmann domains and use this motif to identify all RLM domains in the Protein Data Bank (PDB), thus finding they constitute about 20% of all known 3D structures. The Evolutionary Classification of protein structure Domains (ECOD) classifies RLM domains in a number of groups that lack evidence for homology (X-groups), which suggests that they could have evolved independently multiple times. Closely related, homologous RLM enzyme families can diverge to bind different ligands using similar binding sites and to catalyze different reactions. Conversely, non-homologous RLM domains can converge to catalyze the same reactions or to bind the same ligand with alternate binding modes. We discuss a special case of such convergent evolution that is relevant to the polypharmacology paradigm, wherein the same drug (methotrexate) binds to multiple non-homologous RLM drug targets with different topologies. Finally, assigning proteins with RLM domain to the Enzyme Commission classification suggest that RLM enzymes function mainly in metabolism (and comprise 38% of reference metabolic pathways) and are overrepresented in extant pathways that represent ancient biosynthetic routes such as nucleotide metabolism, energy metabolism, and metabolism of amino acids. In fact, RLM enzymes take part in five out of eight enzymatic reactions of the Wood-Ljungdahl metabolic pathway thought to be used by the last universal common ancestor (LUCA). The prevalence of RLM domains in this ancient metabolism might explain their wide distribution among enzymes.

ECAMI: simultaneous classification and motif identification for enzyme annotation.

Carbohydrate-active enzymes (CAZymes) are extremely important to bioenergy, human gut microbiome, and plant pathogen researches and industries. Here we developed a new amino acid k-mer-based CAZyme classification, motif identification and genome annotation tool using a bipartite network algorithm. Using this tool, we classified 390 CAZyme families into thousands of subfamilies each with distinguishing k-mer peptides. These k-mers represented the characteristic motifs (in the form of a collection of conserved short peptides) of each subfamily, and thus were further used to annotate new genomes for CAZymes. This idea was also generalized to extract characteristic k-mer peptides for all the Swiss-Prot enzymes classified by the EC (enzyme commission) numbers and applied to enzyme EC prediction. This new tool was implemented as a Python package named eCAMI. Benchmark analysis of eCAMI against the state-of-the-art tools on CAZyme and enzyme EC datasets found that: (i) eCAMI has the best performance in terms of accuracy and memory use for CAZyme and enzyme EC classification and annotation; (ii) the k-mer-based tools (including PPR-Hotpep, CUPP and eCAMI) perform better than homology-based tools and deep-learning tools in enzyme EC prediction. Lastly, we confirmed that the k-mer-based tools have the unique ability to identify the characteristic k-mer peptides in the predicted enzymes. https://github.com/yinlabniu/eCAMI and https://github.com/zhanglabNKU/eCAMI. Supplementary data are available at Bioinformatics online.

Challenges in the annotation of pseudoenzymes in databases: the UniProtKB approach.

The universal protein knowledgebase (UniProtKB) collects and centralises functional information on proteins across a wide range of species. In addition to the functional information added to all protein entries, for enzymes, which represent 20-40% of most proteomes, UniProtKB provides additional information about Enzyme Commission classification, catalytic activity, cofactors, enzyme regulation, kinetics and pathways, all based on critical assessment of published experimental data. Computer-based analysis and structural data are used to enrich the annotation of the sequence through the identification of active sites and binding sites. While the annotation of enzymes is well-defined, the curation of pseudoenzymes in UniProtKB has highlighted some challenges: how to identify them, how to assess their lack of catalytic activity, how to annotate their lack of catalytic activity in a consistent way and how much can be inferred and propagated from experimental data obtained from other species. Through various examples, we illustrate some of these issues and discuss some of the changes we propose to enhance the annotation and discovery of pseudoenzymes. Ultimately, improving the curation of pseudoenzymes will provide the scientific community with a comprehensive resource for pseudoenzymes, which in turn will lead to a better understanding of the evolution of these molecules, the aetiology of related diseases and the development of drugs.

The effect of an oligosaccharide reducing-end xylanase, BhRex8A, on the synergistic degradation of xylan backbones by an optimised xylanolytic enzyme cocktail

Xylan, the most abundant hemicellulose in lignocellulosic biomass, requires a consortium of xylanolytic enzymes to achieve its complete de-polymerisation. As global interest in using xylan-containing lignocellulosic feedstocks for biofuel production increases, an accompanying knowledge on how to efficiently depolymerise these feedstocks into fermentable sugars is required. Since it has been observed that the same enzyme [i.e. an enzyme with the same EC (Enzyme Commission) classification] from different GH families can display different substrate specificities and properties, we evaluated GH10 (XT6) and 11 (Xyn2A) xylanase performance alone, and in combination, during xylan depolymerisation. Synergistic enhancement with respect to reducing sugar release was observed when Xyn2A at 75% loading was supplemented with 25% loading of XT6 for both beechwood glucuronoxylan (1.14-fold improvement) and wheat arabinoxylan (1.1-fold improvement) degradation. Following this, the optimised xylanase mixture was dosed with an oligosaccharide reducing-end xylanase (Rex8A) from either Bifidobacterium adolescentis or Bacillus halodurans for further synergistic enhancement. Dosing 75% of the xylanase mixture (Xyn2A:XT6 at 75:25%) with 25% loading of Rex8A led to an enhancement of reducing sugar (up to an 1.1-fold improvement) and xylose release (up to an 1.5-fold improvement); however, this effect was both xylan and Rex8A specific. Using thin layer chromatography, synergism appeared to be a result of the GH10 and 11 xylanases liberating xylo-oligomers that are preferred substrates of the processive Rex8As. Rex8As then hydrolysed xylo-oligomers to xylose - and xylobiose which was the preferred substrate for xylosidase, SXA. This likely explains why there was a significant improvement in xylose release in the presence of Rex8As. Here, it was shown that Rex8As are key enzymes in the efficient saccharification of hetero-xylan into xylose, a major component of lignocellulosic substrates.

EClerize: A customized force-directed graph drawing algorithm for biological graphs with EC attributes.

Visualizing large-scale data produced by the high throughput experiments as a biological graph leads to better understanding and analysis. This study describes a customized force-directed layout algorithm, EClerize, for biological graphs that represent pathways in which the nodes are associated with Enzyme Commission (EC) attributes. The nodes with the same EC class numbers are treated as members of the same cluster. Positions of nodes are then determined based on both the biological similarity and the connection structure. EClerize minimizes the intra-cluster distance, that is the distance between the nodes of the same EC cluster and maximizes the inter-cluster distance, that is the distance between two distinct EC clusters. EClerize is tested on a number of biological pathways and the improvement brought in is presented with respect to the original algorithm. EClerize is available as a plug-in to Cytoscape ( http://apps.cytoscape.org/apps/eclerize ).

Similarity in Shape Dictates Signature Intrinsic Dynamics Despite No Functional Conservation in TIM Barrel Enzymes.

The conservation of the intrinsic dynamics of proteins emerges as we attempt to understand the relationship between sequence, structure and functional conservation. We characterise the conservation of such dynamics in a case where the structure is conserved but function differs greatly. The triosephosphate isomerase barrel fold (TBF), renowned for its 8 β-strand-α-helix repeats that close to form a barrel, is one of the most diverse and abundant folds found in known protein structures. Proteins with this fold have diverse enzymatic functions spanning five of six Enzyme Commission classes, and we have picked five different superfamily candidates for our analysis using elastic network models. We find that the overall shape is a large determinant in the similarity of the intrinsic dynamics, regardless of function. In particular, the β-barrel core is highly rigid, while the α-helices that flank the β-strands have greater relative mobility, allowing for the many possibilities for placement of catalytic residues. We find that these elements correlate with each other via the loops that link them, as opposed to being directly correlated. We are also able to analyse the types of motions encoded by the normal mode vectors of the α-helices. We suggest that the global conservation of the intrinsic dynamics in the TBF contributes greatly to its success as an enzymatic scaffold both through evolution and enzyme design.

MINEs: open access databases of computationally predicted enzyme promiscuity products for untargeted metabolomics

BackgroundIn spite of its great promise, metabolomics has proven difficult to execute in an untargeted and generalizable manner. Liquid chromatography–mass spectrometry (LC–MS) has made it possible to gather data on thousands of cellular metabolites. However, matching metabolites to their spectral features continues to be a bottleneck, meaning that much of the collected information remains uninterpreted and that new metabolites are seldom discovered in untargeted studies. These challenges require new approaches that consider compounds beyond those available in curated biochemistry databases.DescriptionHere we present Metabolic In silico Network Expansions (MINEs), an extension of known metabolite databases to include molecules that have not been observed, but are likely to occur based on known metabolites and common biochemical reactions. We utilize an algorithm called the Biochemical Network Integrated Computational Explorer (BNICE) and expert-curated reaction rules based on the Enzyme Commission classification system to propose the novel chemical structures and reactions that comprise MINE databases. Starting from the Kyoto Encyclopedia of Genes and Genomes (KEGG) COMPOUND database, the MINE contains over 571,000 compounds, of which 93% are not present in the PubChem database. However, these MINE compounds have on average higher structural similarity to natural products than compounds from KEGG or PubChem. MINE databases were able to propose annotations for 98.6% of a set of 667 MassBank spectra, 14% more than KEGG alone and equivalent to PubChem while returning far fewer candidates per spectra than PubChem (46 vs. 1715 median candidates). Application of MINEs to LC–MS accurate mass data enabled the identity of an unknown peak to be confidently predicted.ConclusionsMINE databases are freely accessible for non-commercial use via user-friendly web-tools at http://minedatabase.mcs.anl.gov and developer-friendly APIs. MINEs improve metabolomics peak identification as compared to general chemical databases whose results include irrelevant synthetic compounds. Furthermore, MINEs complement and expand on previous in silico generated compound databases that focus on human metabolism. We are actively developing the database; future versions of this resource will incorporate transformation rules for spontaneous chemical reactions and more advanced filtering and prioritization of candidate structures.Graphical abstractMINE database construction and access methods. The process of constructing a MINE database from the curated source databases is depicted on the left. The methods for accessing the database are shown on the right.Electronic supplementary materialThe online version of this article (doi:10.1186/s13321-015-0087-1) contains supplementary material, which is available to authorized users.

Estimation of protein function using template-based alignment of enzyme active sites

BackgroundThe accumulation of protein structural data occurs more rapidly than it can be characterized by traditional laboratory means. This has motivated widespread efforts to predict enzyme function computationally. The most useful/accurate strategies employed to date are based on the detection of motifs in novel structures that correspond to a specific function. Functional residues are critical components of predictively useful motifs. We have implemented a novel method, to complement current approaches, which detects motifs solely on the basis of distance restraints between catalytic residues.ResultsProMOL is a plugin for the PyMOL molecular graphics environment that can be used to create active site motifs for enzymes. A library of 181 active site motifs has been created with ProMOL, based on definitions published in the Catalytic Site Atlas (CSA). Searches with ProMOL produce better than 50% useful Enzyme Commission (EC) class suggestions for level 1 searches in EC classes 1, 4 and 5, and produce some useful results for other classes. 261 additional motifs automatically translated from Jonathan Barker’s JESS motif set [Bioinformatics 19:1644–1649, 2003] and a set of NMR motifs is under development. Alignments are evaluated by visual superposition, Levenshtein distance and root-mean-square deviation (RMSD) and are reasonably consistent with related search methods.ConclusionThe ProMOL plugin for PyMOL provides ready access to template-based local alignments. Recent improvements to ProMOL, including the expanded motif library, RMSD calculations and output selection formatting, have greatly increased the program’s usability and speed, and have improved the way that the results are presented.

Systematic Detection of Internal Symmetry in Proteins Using CE-Symm

Symmetry is an important feature of protein tertiary and quaternary structures that has been associated with protein folding, function, evolution, and stability. Its emergence and ensuing prevalence has been attributed to gene duplications, fusion events, and subsequent evolutionary drift in sequence. This process maintains structural similarity and is further supported by this study. To further investigate the question of how internal symmetry evolved, how symmetry and function are related, and the overall frequency of internal symmetry, we developed an algorithm, CE-Symm, to detect pseudo-symmetry within the tertiary structure of protein chains. Using a large manually curated benchmark of 1007 protein domains, we show that CE-Symm performs significantly better than previous approaches. We use CE-Symm to build a census of symmetry among domain superfamilies in SCOP and note that 18% of all superfamilies are pseudo-symmetric. Our results indicate that more domains are pseudo-symmetric than previously estimated. We establish a number of recurring types of symmetry–function relationships and describe several characteristic cases in detail. With the use of the Enzyme Commission classification, symmetry was found to be enriched in some enzyme classes but depleted in others. CE-Symm thus provides a methodology for a more complete and detailed study of the role of symmetry in tertiary protein structure [availability: CE-Symm can be run from the Web at http://source.rcsb.org/jfatcatserver/symmetry.jsp. Source code and software binaries are also available under the GNU Lesser General Public License (version 2.1) at https://github.com/rcsb/symmetry. An interactive census of domains identified as symmetric by CE-Symm is available from http://source.rcsb.org/jfatcatserver/scopResults.jsp].

Exploring the Biological and Chemical Complexity of the Ligases

Using a novel method to map and cluster chemical reactions, we have re-examined the chemistry of the ligases [Enzyme Commission (EC) Class 6] and their associated protein families in detail. The type of bond formed by the ligase can be automatically extracted from the equation of the reaction, replicating the EC subclass division. However, this subclass division hides considerable complexities, especially for the C–N forming ligases, which fall into at least three distinct types. The lower levels of the EC classification for ligases are somewhat arbitrary in their definition and add little to understanding their chemistry or evolution. By comparing the multi-domain architecture of the enzymes and using sequence similarity networks, we examined the links between overall reaction and evolution of the ligases. These show that, whilst many enzymes that perform the same overall chemistry group together, both convergent (similar function, different ancestral lineage) and divergent (different function, common ancestor) evolution of function are observed. However, a common theme is that a single conserved domain (often the nucleoside triphosphate binding domain) is combined with ancillary domains that provide the variation in substrate binding and function.

Fifty‐five years of enzyme classification: advances and difficulties

Since the publication of a list of enzymes classified according to the reactions that they catalysed, by Dixon and Webb in 1958, its content and presentation have undergone a number of significant changes. These have been necessitated by new information, as well as the need to improve clarity. The move from printed versions to the online environment, through the ExplorEnz website, has allowed the process of adding newly reported enzymes to be automated and the information content to be enriched. Search and output facilities have also been enhanced. These and the problems attendant on the use of the Enzyme Commission classification system for some groups of enzymes are the subject of this review.

Examining the Flexibility of Tim Barrel Proteins Based on their Structural Topology

The eight-fold beta/alpha barrel-like fold, first seen in triose-phosphate isomerase (TIM), is a common and versatile fold for proteins. Proteins in this classification have diverse enzymatic functions spanning five of six Enzyme Commission classes. Studies have characterized the sequence, structure, localized dynamics and electrostatics of TIM proteins, but the flexibility of the fold is understudied. We use coarse-grained and all-atom normal modes analysis to look at the principle TIM beta/alpha barrel secondary structure elements of five structures from distinct families with low sequence similarity between 23 and 38.5 % and low structure similarity. We report that the eight core beta strands behave independently of the eight flanking alpha helices. The strands and helices repeat in tandem at the sequence level, yet the strands form a core that moves minimally while the helices display greater relative mobility in all of the structures. This core formed by the beta-strands is rigid in all the structures. This is despite the varying loop lengths between them and the presence of accessory secondary structures in some cases. We found that the atomic fluctuations of each of the beta-strands have a high correlation with its immediate beta-strand neighbour in the core. Compared to the beta strands, the alpha helices have less correlation with each other and almost none with the strands. We also defined rotation and translation vectors to decompose the type of movements to which the normal modes of these TIMs correspond. We relate differences in the mobility of the individual secondary structure elements to the differences in function and structure of the five different enzymes.

Relationship between global structural parameters and Enzyme Commission hierarchy: Implications for function prediction

In protein databases there is a substantial number of proteins structurally determined but without function annotation. Understanding the relationship between function and structure can be useful to predict function on a large scale. We have analyzed the similarities in global physicochemical parameters for a set of enzymes which were classified according to the four Enzyme Commission (EC) hierarchical levels. Using relevance theory we introduced a distance between proteins in the space of physicochemical characteristics. This was done by minimizing a cost function of the metric tensor built to reflect the EC classification system. Using an unsupervised clustering method on a set of 1025 enzymes, we obtained no relevant clustering formation compatible with EC classification. The distance distributions between enzymes from the same EC group and from different EC groups were compared by histograms. Such analysis was also performed using sequence alignment similarity as a distance. Our results suggest that global structure parameters are not sufficient to segregate enzymes according to EC hierarchy. This indicates that features essential for function are rather local than global. Consequently, methods for predicting function based on global attributes should not obtain high accuracy in main EC classes prediction without relying on similarities between enzymes from training and validation datasets. Furthermore, these results are consistent with a substantial number of studies suggesting that function evolves fundamentally by recruitment, i.e., a same protein motif or fold can be used to perform different enzymatic functions and a few specific amino acids (AAs) are actually responsible for enzyme activity. These essential amino acids should belong to active sites and an effective method for predicting function should be able to recognize them.

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

BackgroundManual annotation of enzymatic functions cannot keep up with automatic genome sequencing. In this work we explore the capacity of InterPro sequence signatures to automatically predict enzymatic function.ResultsWe present EnzML, a multi-label classification method that can efficiently account also for proteins with multiple enzymatic functions: 50,000 in UniProt. EnzML was evaluated using a standard set of 300,747 proteins for which the manually curated Swiss-Prot and KEGG databases have agreeing Enzyme Commission (EC) annotations. EnzML achieved more than 98% subset accuracy (exact match of all correct Enzyme Commission classes of a protein) for the entire dataset and between 87 and 97% subset accuracy in reannotating eight entire proteomes: human, mouse, rat, mouse-ear cress, fruit fly, the S. pombe yeast, the E. coli bacterium and the M. jannaschii archaebacterium. To understand the role played by the dataset size, we compared the cross-evaluation results of smaller datasets, either constructed at random or from specific taxonomic domains such as archaea, bacteria, fungi, invertebrates, plants and vertebrates. The results were confirmed even when the redundancy in the dataset was reduced using UniRef100, UniRef90 or UniRef50 clusters.ConclusionsInterPro signatures are a compact and powerful attribute space for the prediction of enzymatic function. This representation makes multi-label machine learning feasible in reasonable time (30 minutes to train on 300,747 instances with 10,852 attributes and 2,201 class values) using the Mulan Binary Relevance Nearest Neighbours algorithm implementation (BR-kNN).

Is EC class predictable from reaction mechanism?

BackgroundWe investigate the relationships between the EC (Enzyme Commission) class, the associated chemical reaction, and the reaction mechanism by building predictive models using Support Vector Machine (SVM), Random Forest (RF) and k-Nearest Neighbours (kNN). We consider two ways of encoding the reaction mechanism in descriptors, and also three approaches that encode only the overall chemical reaction. Both cross-validation and also an external test set are used.ResultsThe three descriptor sets encoding overall chemical transformation perform better than the two descriptions of mechanism. SVM and RF models perform comparably well; kNN is less successful. Oxidoreductases and hydrolases are relatively well predicted by all types of descriptor; isomerases are well predicted by overall reaction descriptors but not by mechanistic ones.ConclusionsOur results suggest that pairs of similar enzyme reactions tend to proceed by different mechanisms. Oxidoreductases, hydrolases, and to some extent isomerases and ligases, have clear chemical signatures, making them easier to predict than transferases and lyases. We find evidence that isomerases as a class are notably mechanistically diverse and that their one shared property, of substrate and product being isomers, can arise in various unrelated ways.The performance of the different machine learning algorithms is in line with many cheminformatics applications, with SVM and RF being roughly equally effective. kNN is less successful, given the role that non-local information plays in successful classification. We note also that, despite a lack of clarity in the literature, EC number prediction is not a single problem; the challenge of predicting protein function from available sequence data is quite different from assigning an EC classification from a cheminformatics representation of a reaction.