ECPred: a tool for the prediction of the enzymatic functions of protein sequences based on the EC nomenclature

Abstract

BackgroundThe automated prediction of the enzymatic functions of uncharacterized proteins is a crucial topic in bioinformatics. Although several methods and tools have been proposed to classify enzymes, most of these studies are limited to specific functional classes and levels of the Enzyme Commission (EC) number hierarchy. Besides, most of the previous methods incorporated only a single input feature type, which limits the applicability to the wide functional space. Here, we proposed a novel enzymatic function prediction tool, ECPred, based on ensemble of machine learning classifiers.ResultsIn ECPred, each EC number constituted an individual class and therefore, had an independent learning model. Enzyme vs. non-enzyme classification is incorporated into ECPred along with a hierarchical prediction approach exploiting the tree structure of the EC nomenclature. ECPred provides predictions for 858 EC numbers in total including 6 main classes, 55 subclass classes, 163 sub-subclass classes and 634 substrate classes. The proposed method is tested and compared with the state-of-the-art enzyme function prediction tools by using independent temporal hold-out and no-Pfam datasets constructed during this study.ConclusionsECPred is presented both as a stand-alone and a web based tool to provide probabilistic enzymatic function predictions (at all five levels of EC) for uncharacterized protein sequences. Also, the datasets of this study will be a valuable resource for future benchmarking studies. ECPred is available for download, together with all of the datasets used in this study, at: https://github.com/cansyl/ECPred. ECPred webserver can be accessed through http://cansyl.metu.edu.tr/ECPred.html.

UDSMProt: Universal Deep Sequence Models for Protein Classification

“…If not noted otherwise, CNN models are trained on representative sequences as this considerably reduces the computational burden for determining PSSM features and is in line with the literature, see e.g. [Li et al, 2018, Dalkiran et al, 2018, whereas UDSMProt is conventionally trained using the full training set including redundant sequences, whereas the corresponding test and validation sets always contain only non-redundant sequences. For the EC50 dataset non-redundant sequences enlarge the size of the training set from 45k to 114k and from 86k to 170k sequences for level 1 and level 0 respectively.…”

“…This process is repeated to a desired number of iterations. In our experiments we used the same parameters as reported in the literature [Li et al, 2018, Dalkiran et al, 2018, Shen and Chou, 2007, namely three iterations with e_value = 0.001, where e_value relates to the threshold for which an alignment is considered as significant. While the raw sequences from Swiss-Prot contained 26 unique amino acids (20 standard and 6 non-standard amino acids), PSSM features are computed only for the 20 standard amino acids.…”

“…Instead of training simple classifiers on highly engineered features, they trained feature representation and classification in an end-to-end fashion with a hybrid CNN-LSTM-approach. Recently, ECPred [Dalkiran et al, 2018] also showed competitive results by building an ensemble of well-performing classifiers (Subsequence Profile Map with PSSM [Sarac et al, 2008], BLAST-kNN [Madden, 2013] and Pepstats-SVM using peptides statistics [Rice et al, 2000]). Nevertheless, drawbacks as described in Section 1 remain, i.e.…”

“…In order to relate our proposed approach to state-of-the-art methods in literature, we conducted an experiment on two datasets provided by ECPred [Dalkiran et al, 2018] and DEEPre [Li et al, 2018]. One of the purposes of this analysis is to justify our choice of the CNN baseline algorithm by demonstrating that it performs on par with state-of-the-art algorithms that do not make use of additional side-information e.g.…”

“…There is a large body of literature on methods to infer protein properties, most of which make use of additional handcrafted features in addition to the primary sequence alone [Shen and Chou, 2007, Håndstad et al, 2007, Gong et al, 2016, Cozzetto et al, 2016, Li et al, 2017, Dalkiran et al, 2018. These include experimentally determined functional annotations (such as Pfam [El-Gebali et al, 2018]) as well as features incorporating information from homologous (evolutionary related) proteins that are typically inferred from well-motivated but still heuristic methods such as the basic local alignment search tool (BLAST) [Madden, 2013], that searches a database for proteins that are homologous to a given query protein, via multiple sequence alignment.…”