Functional classification of CATH superfamilies: a domain-based approach for protein function annotation

DeepFunc: A Deep Learning Framework for Accurate Prediction of Protein Functions from Protein Sequences and Interactions

Annotation of protein functions plays an important role in understanding life at the molecular level. High-throughput sequencing produces massive numbers of raw proteins sequences and only about 1% of them have been manually annotated with functions. Experimental annotations of functions are expensive, time-consuming and do not keep up with the rapid growth of the sequence numbers. This motivates the development of computational approaches that predict protein functions. A novel deep learning framework, DeepFunc, is proposed which accurately predicts protein functions from protein sequence- and network-derived information. More precisely, DeepFunc uses a long and sparse binary vector to encode information concerning domains, families, and motifs collected from the InterPro tool that is associated with the input protein sequence. This vector is processed with two neural layers to obtain a low-dimensional vector which is combined with topological information extracted from protein-protein interactions (PPIs) and functional linkages. The combined information is processed by a deep neural network that predicts protein functions. DeepFunc is empirically and comparatively tested on a benchmark testing dataset and the Critical Assessment of protein Function Annotation algorithms (CAFA) 3 dataset. The experimental results demonstrate that DeepFunc outperforms current methods on the testing dataset and that it secures the highest F_max  = 0.54 and AUC = 0.94 on the CAFA3 dataset.

A CATH domain functional family based approach to identify putative cancer driver genes and driver mutations

Tumour sequencing identifies highly recurrent point mutations in cancer driver genes, but rare functional mutations are hard to distinguish from large numbers of passengers. We developed a novel computational platform applying a multi-modal approach to filter out passengers and more robustly identify putative driver genes. The primary filter identifies enrichment of cancer mutations in CATH functional families (CATH-FunFams) – structurally and functionally coherent sets of evolutionary related domains. Using structural representatives from CATH-FunFams, we subsequently seek enrichment of mutations in 3D and show that these mutation clusters have a very significant tendency to lie close to known functional sites or conserved sites predicted using CATH-FunFams. Our third filter identifies enrichment of putative driver genes in functionally coherent protein network modules confirmed by literature analysis to be cancer associated. Our approach is complementary to other domain enrichment approaches exploiting Pfam families, but benefits from more functionally coherent groupings of domains. Using a set of mutations from 22 cancers we detect 151 putative cancer drivers, of which 79 are not listed in cancer resources and include recently validated cancer associated genes EPHA7, DCC netrin-1 receptor and zinc-finger protein ZNF479.

CATH: expanding the horizons of structure-based functional annotations for genome sequences

This article provides an update of the latest data and developments within the CATH protein structure classification database (http://www.cathdb.info). The resource provides two levels of release: CATH-B, a daily snapshot of the latest structural domain boundaries and superfamily assignments, and CATH+, which adds layers of derived data, such as predicted sequence domains, functional annotations and functional clustering (known as Functional Families or FunFams). The most recent CATH+ release (version 4.2) provides a huge update in the coverage of structural data. This release increases the number of fully- classified domains by over 40% (from 308 999 to 434 857 structural domains), corresponding to an almost two- fold increase in sequence data (from 53 million to over 95 million predicted domains) organised into 6119 superfamilies. The coverage of high-resolution, protein PDB chains that contain at least one assigned CATH domain is now 90.2% (increased from 82.3% in the previous release). A number of highly requested features have also been implemented in our web pages: allowing the user to view an alignment between their query sequence and a representative FunFam structure and providing tools that make it easier to view the full structural context (multi-domain architecture) of domains and chains.

DeepGO: predicting protein functions from sequence and interactions using a deep ontology-aware classifier

Motivation

A large number of protein sequences are becoming available through the application of novel high-throughput sequencing technologies. Experimental functional characterization of these proteins is time-consuming and expensive, and is often only done rigorously for few selected model organisms. Computational function prediction approaches have been suggested to fill this gap. The functions of proteins are classified using the Gene Ontology (GO), which contains over 40 000 classes. Additionally, proteins have multiple functions, making function prediction a large-scale, multi-class, multi-label problem.

Results

We have developed a novel method to predict protein function from sequence. We use deep learning to learn features from protein sequences as well as a cross-species protein–protein interaction network. Our approach specifically outputs information in the structure of the GO and utilizes the dependencies between GO classes as background information to construct a deep learning model. We evaluate our method using the standards established by the Computational Assessment of Function Annotation (CAFA) and demonstrate a significant improvement over baseline methods such as BLAST, in particular for predicting cellular locations.

Availability and implementation

Web server: http://deepgo.bio2vec.net, Source code: https://github.com/bio-ontology-research-group/deepgo

Supplementary information

Supplementary data are available at Bioinformatics online.

Protein CoAlation and antioxidant function of coenzyme A in prokaryotic cells

In all living organisms, coenzyme A (CoA) is an essential cofactor with a unique design allowing it to function as an acyl group carrier and a carbonyl-activating group in diverse biochemical reactions. It is synthesized in a highly conserved process in prokaryotes and eukaryotes that requires pantothenic acid (vitamin B5), cysteine and ATP. CoA and its thioester derivatives are involved in major metabolic pathways, allosteric interactions and the regulation of gene expression. A novel unconventional function of CoA in redox regulation has been recently discovered in mammalian cells and termed protein CoAlation. Here, we report for the first time that protein CoAlation occurs at a background level in exponentially growing bacteria and is strongly induced in response to oxidizing agents and metabolic stress. Over 12% of Staphylococcus aureus gene products were shown to be CoAlated in response to diamide-induced stress. In vitro CoAlation of S. aureus glyceraldehyde-3-phosphate dehydrogenase was found to inhibit its enzymatic activity and to protect the catalytic cysteine 151 from overoxidation by hydrogen peroxide. These findings suggest that in exponentially growing bacteria, CoA functions to generate metabolically active thioesters, while it also has the potential to act as a low-molecular-weight antioxidant in response to oxidative and metabolic stress.

HMMER Cut-off Threshold Tool (HMMERCTTER): Supervised classification of superfamily protein sequences with a reliable cut-off threshold

Background

Protein superfamilies can be divided into subfamilies of proteins with different functional characteristics. Their sequences can be classified hierarchically, which is part of sequence function assignation. Typically, there are no clear subfamily hallmarks that would allow pattern-based function assignation by which this task is mostly achieved based on the similarity principle. This is hampered by the lack of a score cut-off that is both sensitive and specific.

Results

HMMER Cut-off Threshold Tool (HMMERCTTER) adds a reliable cut-off threshold to the popular HMMER. Using a high quality superfamily phylogeny, it clusters a set of training sequences such that the cluster-specific HMMER profiles show cluster or subfamily member detection with 100% precision and recall (P&R), thereby generating a specific threshold as inclusion cut-off. Profiles and thresholds are then used as classifiers to screen a target dataset. Iterative inclusion of novel sequences to groups and the corresponding HMMER profiles results in high sensitivity while specificity is maintained by imposing 100% P&R self detection. In three presented case studies of protein superfamilies, classification of large datasets with 100% precision was achieved with over 95% recall. Limits and caveats are presented and explained.

Conclusions

HMMERCTTER is a promising protein superfamily sequence classifier provided high quality training datasets are used. It provides a decision support system that aids in the difficult task of sequence function assignation in the twilight zone of sequence similarity. All relevant data and source codes are available from the Github repository at the following URL: https://github.com/BBCMdP/HMMERCTTER.

The Classification of Protein Domains

The significant expansion in protein sequence and structure data that we are now witnessing brings with it a pressing need to bring order to the protein world. Such order enables us to gain insights into the evolution of proteins, their function and the extent to which the functional repertoire can vary across the three kingdoms of life. This has lead to the creation of a wide range of protein family classifications that aim to group proteins based upon their evolutionary relationships.In this chapter we discuss the approaches and methods that are frequently used in the classification of proteins, with a specific emphasis on the classification of protein domains. The construction of both domain sequence and domain structure databases is considered and we show how the use of domain family annotations to assign structural and functional information is enhancing our understanding of genomes.

Large-scale automated function prediction of protein sequences and an experimental case study validation on PTEN transcript variants

Recent advances in computing power and machine learning empower functional annotation of protein sequences and their transcript variations. Here, we present an automated prediction system UniGOPred, for GO annotations and a database of GO term predictions for proteomes of several organisms in UniProt Knowledgebase (UniProtKB). UniGOPred provides function predictions for 514 molecular function (MF), 2909 biological process (BP), and 438 cellular component (CC) GO terms for each protein sequence. UniGOPred covers nearly the whole functionality spectrum in Gene Ontology system and it can predict both generic and specific GO terms. UniGOPred was run on CAFA2 challenge target protein sequences and it is categorized within the top 10 best performing methods for the molecular function category. In addition, the performance of UniGOPred is higher compared to the baseline BLAST classifier in all categories of GO. UniGOPred predictions are compared with UniProtKB/TrEMBL database annotations as well. Furthermore, the proposed tool's ability to predict negatively associated GO terms that defines the functions that a protein does not possess, is discussed. UniGOPred annotations were also validated by case studies on PTEN protein variants experimentally and on CHD8 protein variants with literature. UniGOPred protein functional annotation system is available as an open access tool at http://cansyl.metu.edu.tr/UniGOPred.html.

PixelDB: Protein-peptide complexes annotated with structural conservation of the peptide binding mode

PixelDB, the Peptide Exosite Location Database, compiles 1966 non-redundant, high-resolution structures of protein-peptide complexes filtered to minimize the impact of crystal packing on peptide conformation. The database is organized to facilitate study of structurally conserved versus non-conserved elements of protein-peptide engagement. PixelDB clusters complexes based on the structural similarity of the peptide-binding protein, and by comparing complexes within a cluster highlights examples of domains that engage peptides using more than one binding mode. PixelDB also identifies conserved peptide core structural motifs characteristic of each binding mode. Peptide regions that flank core motifs often make non-structurally conserved interactions with the protein surface in regions we call exosites. Many examples establish that exosite contacts can be important for enhancing protein binding and interaction specificity. PixelDB provides a resource for computational and structural biologists to study, model, and predict core-motif and exosite-contacting peptide interactions. PixelDB is available to the community without restriction in a convenient flat-file format with accompanying visualization tools.

Enzyme function and its evolution

With rapid increases over recent years in the determination of protein sequence and structure, alongside knowledge of thousands of enzyme functions and hundreds of chemical mechanisms, it is now possible to combine breadth and depth in our understanding of enzyme evolution. Phylogenetics continues to move forward, though determining correct evolutionary family trees is not trivial. Protein function prediction has spawned a variety of promising methods that offer the prospect of identifying enzymes across the whole range of chemical functions and over numerous species. This knowledge is essential to understand antibiotic resistance, as well as in protein re-engineering and de novo enzyme design.

IntPred: a structure-based predictor of protein-protein interaction sites

Motivation

Protein–protein interactions are vital for protein function with the average protein having between three and ten interacting partners. Knowledge of precise protein–protein interfaces comes from crystal structures deposited in the Protein Data Bank (PDB), but only 50% of structures in the PDB are complexes. There is therefore a need to predict protein–protein interfaces in silico and various methods for this purpose. Here we explore the use of a predictor based on structural features and which exploits random forest machine learning, comparing its performance with a number of popular established methods.

Results

On an independent test set of obligate and transient complexes, our IntPred predictor performs well (MCC = 0.370, ACC = 0.811, SPEC = 0.916, SENS = 0.411) and compares favourably with other methods. Overall, IntPred ranks second of six methods tested with SPPIDER having slightly better overall performance (MCC = 0.410, ACC = 0.759, SPEC = 0.783, SENS = 0.676), but considerably worse specificity than IntPred. As with SPPIDER, using an independent test set of obligate complexes enhanced performance (MCC = 0.381) while performance is somewhat reduced on a dataset of transient complexes (MCC = 0.303). The trade-off between sensitivity and specificity compared with SPPIDER suggests that the choice of the appropriate tool is application-dependent.

Availability and implementation

IntPred is implemented in Perl and may be downloaded for local use or run via a web server at www.bioinf.org.uk/intpred/.

Supplementary information

Supplementary data are available at Bioinformatics online.

Structural and Functional View of Polypharmacology

Protein domains mediate drug-protein interactions and this principle can guide the design of multi-target drugs i.e. polypharmacology. In this study, we associate multi-target drugs with CATH functional families through the overrepresentation of targets of those drugs in CATH functional families. Thus, we identify CATH functional families that are currently enriched in drugs (druggable CATH functional families) and we use the network properties of these druggable protein families to analyse their association with drug side effects. Analysis of selected druggable CATH functional families, enriched in drug targets, show that relatives exhibit highly conserved drug binding sites. Furthermore, relatives within druggable CATH functional families occupy central positions in a human protein functional network, cluster together forming network neighbourhoods and are less likely to be within proteins associated with drug side effects. Our results demonstrate that CATH functional families can be used to identify drug-target interactions, opening a new research direction in target identification.

An overview of comparative modelling and resources dedicated to large-scale modelling of genome sequences

Computational modelling of proteins has been a major catalyst in structural biology. Bioinformatics groups have exploited the repositories of known structures to predict high-quality structural models with high efficiency at low cost. This article provides an overview of comparative modelling, reviews recent developments and describes resources dedicated to large-scale comparative modelling of genome sequences. The value of subclustering protein domain superfamilies to guide the template-selection process is investigated. Some recent cases in which structural modelling has aided experimental work to determine very large macromolecular complexes are also cited.

CATH-Gene3D: Generation of the Resource and Its Use in Obtaining Structural and Functional Annotations for Protein Sequences

This chapter describes the generation of the data in the CATH-Gene3D online resource and how it can be used to study protein domains and their evolutionary relationships. Methods will be presented for: comparing protein structures, recognizing homologs, predicting domain structures within protein sequences, and subclassifying superfamilies into functionally pure families, together with a guide on using the webpages.

CATH: an expanded resource to predict protein function through structure and sequence

The latest version of the CATH-Gene3D protein structure classification database has recently been released (version 4.1, http://www.cathdb.info). The resource comprises over 300 000 domain structures and over 53 million protein domains classified into 2737 homologous superfamilies, doubling the number of predicted protein domains in the previous version. The daily-updated CATH-B, which contains our very latest domain assignment data, provides putative classifications for over 100 000 additional protein domains. This article describes developments to the CATH-Gene3D resource over the last two years since the publication in 2015, including: significant increases to our structural and sequence coverage; expansion of the functional families in CATH; building a support vector machine (SVM) to automatically assign domains to superfamilies; improved search facilities to return alignments of query sequences against multiple sequence alignments; the redesign of the web pages and download site.

Novel Computational Protocols for Functionally Classifying and Characterising Serine Beta-Lactamases

Beta-lactamases represent the main bacterial mechanism of resistance to beta-lactam antibiotics and are a significant challenge to modern medicine. We have developed an automated classification and analysis protocol that exploits structure- and sequence-based approaches and which allows us to propose a grouping of serine beta-lactamases that more consistently captures and rationalizes the existing three classification schemes: Classes, (A, C and D, which vary in their implementation of the mechanism of action); Types (that largely reflect evolutionary distance measured by sequence similarity); and Variant groups (which largely correspond with the Bush-Jacoby clinical groups). Our analysis platform exploits a suite of in-house and public tools to identify Functional Determinants (FDs), i.e. residue sites, responsible for conferring different phenotypes between different classes, different types and different variants. We focused on Class A beta-lactamases, the most highly populated and clinically relevant class, to identify FDs implicated in the distinct phenotypes associated with different Class A Types and Variants. We show that our FunFHMMer method can separate the known beta-lactamase classes and identify those positions likely to be responsible for the different implementations of the mechanism of action in these enzymes. Two novel algorithms, ASSP and SSPA, allow detection of FD sites likely to contribute to the broadening of the substrate profiles. Using our approaches, we recognise 151 Class A types in UniProt. Finally, we used our beta-lactamase FunFams and ASSP profiles to detect 4 novel Class A types in microbiome samples. Our platforms have been validated by literature studies, in silico analysis and some targeted experimental verification. Although developed for the serine beta-lactamases they could be used to classify and analyse any diverse protein superfamily where sub-families have diverged over both long and short evolutionary timescales.

Gene3D: expanding the utility of domain assignments

Gene3D http://gene3d.biochem.ucl.ac.uk is a database of domain annotations of Ensembl and UniProtKB protein sequences. Domains are predicted using a library of profile HMMs representing 2737 CATH superfamilies. Gene3D has previously featured in the Database issue of NAR and here we report updates to the website and database. The current Gene3D (v14) release has expanded its domain assignments to ∼ 20,000 cellular genomes and over 43 million unique protein sequences, more than doubling the number of protein sequences since our last publication. Amongst other updates, we have improved our Functional Family annotation method. We have also improved the quality and coverage of our 3D homology modelling pipeline of predicted CATH domains. Additionally, the structural models have been expanded to include an extra model organism (Drosophila melanogaster). We also document a number of additional visualization tools in the Gene3D website.

Protein Structure Databases

Web-based protein structure databases come in a wide variety of types and levels of information content. Those having the most general interest are the various atlases that describe each experimentally determined protein structure and provide useful links, analyses, and schematic diagrams relating to its 3D structure and biological function. Also of great interest are the databases that classify 3D structures by their folds as these can reveal evolutionary relationships which may be hard to detect from sequence comparison alone. Related to these are the numerous servers that compare folds-particularly useful for newly solved structures, and especially those of unknown function. Beyond these are a vast number of databases for the more specialized user, dealing with specific families, diseases, structural features, and so on.

Large-Scale Analysis Exploring Evolution of Catalytic Machineries and Mechanisms in Enzyme Superfamilies

Enzymes, as biological catalysts, form the basis of all forms of life. How these proteins have evolved their functions remains a fundamental question in biology. Over 100 years of detailed biochemistry studies, combined with the large volumes of sequence and protein structural data now available, means that we are able to perform large-scale analyses to address this question. Using a range of computational tools and resources, we have compiled information on all experimentally annotated changes in enzyme function within 379 structurally defined protein domain superfamilies, linking the changes observed in functions during evolution to changes in reaction chemistry. Many superfamilies show changes in function at some level, although one function often dominates one superfamily. We use quantitative measures of changes in reaction chemistry to reveal the various types of chemical changes occurring during evolution and to exemplify these by detailed examples. Additionally, we use structural information of the enzymes active site to examine how different superfamilies have changed their catalytic machinery during evolution. Some superfamilies have changed the reactions they perform without changing catalytic machinery. In others, large changes of enzyme function, in terms of both overall chemistry and substrate specificity, have been brought about by significant changes in catalytic machinery. Interestingly, in some superfamilies, relatives perform similar functions but with different catalytic machineries. This analysis highlights characteristics of functional evolution across a wide range of superfamilies, providing insights that will be useful in predicting the function of uncharacterised sequences and the design of new synthetic enzymes.

•

Examining how enzyme function evolves using sequence, structure, and reaction mechanism data.

•

Quantifying changes in reaction mechanisms reveals how function has diverged in many superfamilies.

•

Homologous domains frequently use different catalytic residues, which sometimes perform the same enzyme chemistry.

•

This large-scale analysis has significance in protein function prediction and enzyme design.

Diversity in protein domain superfamilies

Whilst ∼93% of domain superfamilies appear to be relatively structurally and functionally conserved based on the available data from the CATH-Gene3D domain classification resource, the remainder are much more diverse. In this review, we consider how domains in some of the most ubiquitous and promiscuous superfamilies have evolved, in particular the plasticity in their functional sites and surfaces which expands the repertoire of molecules they interact with and actions performed on them. To what extent can we identify a core function for these superfamilies which would allow us to develop a ‘domain grammar of function’ whereby a protein's biological role can be proposed from its constituent domains? Clearly the first step is to understand the extent to which these components vary and how changes in their molecular make-up modifies function.

The history of the CATH structural classification of protein domains

This article presents a historical review of the protein structure classification database CATH. Together with the SCOP database, CATH remains comprehensive and reasonably up-to-date with the now more than 100,000 protein structures in the PDB. We review the expansion of the CATH and SCOP resources to capture predicted domain structures in the genome sequence data and to provide information on the likely functions of proteins mediated by their constituent domains. The establishment of comprehensive function annotation resources has also meant that domain families can be functionally annotated allowing insights into functional divergence and evolution within protein families.

•

We present a historical review of the protein structure database CATH.

•

We review the expansion of the CATH and SCOP resources with sequence data and functional annotations.

•

How functional annotation resources allow insights into functional divergence and evolution within protein families.