MOTIVATION: The Encyclopedia of Domains (TED) provides domain annotations for proteins in the AlphaFold Protein Structure Database (AFDB) using a consensus of three state-of-the-art structure-based methods. We used these annotations to construct profile Hidden Markov models (HMMs), collectively forming the TED Library of HMMs (TEDLH). TEDLH enables sensitive sequence and profile searches, supporting systematic exploration of protein domain families and their evolutionary relationships. RESULTS: TEDLH links 934,186 domain HMMs to experimentally determined CATH-PDB structures through direct (primary) and transitive (secondary and tertiary) relationships. Fewer than half of TEDLH HMMs are directly linked to a CATH-PDB domain; the remaining models are connected through transitive relationships. These transitive links extend coverage into more divergent regions of sequence space and better represent CATH superfamily diversity. HMM-HMM comparisons within CATH superfamily 3.30.70.100 illustrate how transitive relationships expand sequence coverage. In this superfamily, 5640 TEDLH HMMs are connected to 173 CATH-PDB representatives. Primary, secondary, and tertiary relationships progressively capture more divergent sequences (pairwise sequence identity <20%) that retain structural similarity (TM-score ≥0.6) and a conserved two-layer α/β sandwich core fold. All-against-all HMM-HMM comparisons across TEDLH also reveal sequence similarities across the CATH hierarchy (cross-hits). At low query coverage (<50%), cross-hits are more frequent between CATH classes, architectures and topologies, whereas at higher coverage thresholds (≥70%) they predominantly occur between superfamilies. These cross-hits are not driven by superfamily size or sequence diversity and can provide guidance for CATH curation. As an example, analysis of cross-hits between superfamilies 2.170.130.30 and 3.10.20.30 reveals evolutionary relationships between these groups. AVAILABILITY: TEDLH is compatible with HH-suite3 and is available from FigShare https://doi.org/10.6084/m9.figshare.28531754 for local use.

MOTIVATION: Protein domains are fundamental units of protein structure and play a pivotal role in understanding folding, function, evolution, and design. The advent of accurate structure prediction techniques has resulted in an influx of new structural data, making the partitioning of these structures into domains essential for inferring evolutionary relationships and functional classification. RESULTS: This article presents Chainsaw, a supervised learning approach to domain parsing that achieves accuracy that surpasses current state-of-the-art methods. Chainsaw uses a fully convolutional neural network which is trained to predict the probability that each pair of residues is in the same domain. Domain predictions are then derived from these pairwise predictions using an algorithm that searches for the most likely assignment of residues to domains given the set of pairwise co-membership probabilities. Chainsaw matches CATH domain annotations in 78% of protein domains versus 72% for the next closest method. When predicting on AlphaFold models, expert human evaluators were twice as likely to prefer Chainsaw's predictions versus the next best method. AVAILABILITY AND IMPLEMENTATION: github.com/JudeWells/Chainsaw.

MOTIVATION: CATH is a protein domain classification resource that exploits an automated workflow of structure and sequence comparison alongside expert manual curation to construct a hierarchical classification of evolutionary and structural relationships. The aim of this study was to develop algorithms for detecting remote homologues missed by state-of-the-art hidden Markov model (HMM)-based approaches. The method developed (CATHe) combines a neural network with sequence representations obtained from protein language models. It was assessed using a dataset of remote homologues having less than 20% sequence identity to any domain in the training set. RESULTS: The CATHe models trained on 1773 largest and 50 largest CATH superfamilies had an accuracy of 85.6 ± 0.4% and 98.2 ± 0.3%, respectively. As a further test of the power of CATHe to detect more remote homologues missed by HMMs derived from CATH domains, we used a dataset consisting of protein domains that had annotations in Pfam, but not in CATH. By using highly reliable CATHe predictions (expected error rate <0.5%), we were able to provide CATH annotations for 4.62 million Pfam domains. For a subset of these domains from Homo sapiens, we structurally validated 90.86% of the predictions by comparing their corresponding AlphaFold2 structures with structures from the CATH superfamilies to which they were assigned. AVAILABILITY AND IMPLEMENTATION: The code for the developed models is available on https://github.com/vam-sin/CATHe, and the datasets developed in this study can be accessed on https://zenodo.org/record/6327572. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Mutations in human proteins lead to diseases. The structure of these proteins can help understand the mechanism of such diseases and develop therapeutics against them. With improved deep learning techniques, such as RoseTTAFold and AlphaFold, we can predict the structure of proteins even in the absence of structural homologs. We modeled and extracted the domains from 553 disease-associated human proteins without known protein structures or close homologs in the Protein Databank. We noticed that the model quality was higher and the Root mean square deviation (RMSD) lower between AlphaFold and RoseTTAFold models for domains that could be assigned to CATH families as compared to those which could only be assigned to Pfam families of unknown structure or could not be assigned to either. We predicted ligand-binding sites, protein-protein interfaces and conserved residues in these predicted structures. We then explored whether the disease-associated missense mutations were in the proximity of these predicted functional sites, whether they destabilized the protein structure based on ddG calculations or whether they were predicted to be pathogenic. We could explain 80% of these disease-associated mutations based on proximity to functional sites, structural destabilization or pathogenicity. When compared to polymorphisms, a larger percentage of disease-associated missense mutations were buried, closer to predicted functional sites, predicted as destabilizing and pathogenic. Usage of models from the two state-of-the-art techniques provide better confidence in our predictions, and we explain 93 additional mutations based on RoseTTAFold models which could not be explained based solely on AlphaFold models.

BACKGROUND: Protein function prediction remains a key challenge. Domain composition affects protein function. Here we present DomFun, a Ruby gem that uses associations between protein domains and functions, calculated using multiple indices based on tripartite network analysis. These domain-function associations are combined at the protein level, to generate protein-function predictions. RESULTS: We analysed 16 tripartite networks connecting homologous superfamily and FunFam domains from CATH-Gene3D with functional annotations from the three Gene Ontology (GO) sub-ontologies, KEGG, and Reactome. We validated the results using the CAFA 3 benchmark platform for GO annotation, finding that out of the multiple association metrics and domain datasets tested, Simpson index for FunFam domain-function associations combined with Stouffer's method leads to the best performance in almost all scenarios. We also found that using FunFams led to better performance than superfamilies, and better results were found for GO molecular function compared to GO biological process terms. DomFun performed as well as the highest-performing method in certain CAFA 3 evaluation procedures in terms of [Formula: see text] and [Formula: see text] We also implemented our own benchmark procedure, Pathway Prediction Performance (PPP), which can be used to validate function prediction for additional annotations sources, such as KEGG and Reactome. Using PPP, we found similar results to those found with CAFA 3 for GO, moreover we found good performance for the other annotation sources. As with CAFA 3, Simpson index with Stouffer's method led to the top performance in almost all scenarios. CONCLUSIONS: DomFun shows competitive performance with other methods evaluated in CAFA 3 when predicting proteins function with GO, although results vary depending on the evaluation procedure. Through our own benchmark procedure, PPP, we have shown it can also make accurate predictions for KEGG and Reactome. It performs best when using FunFams, combining Simpson index derived domain-function associations using Stouffer's method. The tool has been implemented so that it can be easily adapted to incorporate other protein features, such as domain data from other sources, amino acid k-mers and motifs. The DomFun Ruby gem is available from https://rubygems.org/gems/DomFun . Code maintained at https://github.com/ElenaRojano/DomFun . Validation procedure scripts can be found at https://github.com/ElenaRojano/DomFun_project .

Peripheral membrane proteins (PMPs) include a wide variety of proteins that have in common to bind transiently to the chemically complex interfacial region of membranes through their interfacial binding site (IBS). In contrast to protein-protein or protein-DNA/RNA interfaces, peripheral protein-membrane interfaces are poorly characterized. We collected a dataset of PMP domains representative of the variety of PMP functions: membrane-targeting domains (Annexin, C1, C2, discoidin C2, PH, PX), enzymes (PLA, PLC/D) and lipid-transfer proteins (START). The dataset contains 1328 experimental structures and 1194 AphaFold models. We mapped the amino acid composition and structural patterns of the IBS of each protein in this dataset, and evaluated which were more likely to be found at the IBS compared to the rest of the domains' accessible surface. In agreement with earlier work we find that about two thirds of the PMPs in the dataset have protruding hydrophobes (Leu, Ile, Phe, Tyr, Trp and Met) at their IBS. The three aromatic amino acids Trp, Tyr and Phe are a hallmark of PMPs IBS regardless of whether they protrude on loops or not. This is also the case for lysines but not arginines suggesting that, unlike for Arg-rich membrane-active peptides, the less membrane-disruptive lysine is preferred in PMPs. Another striking observation was the over-representation of glycines at the IBS of PMPs compared to the rest of their surface, possibly procuring IBS loops a much-needed flexibility to insert in-between membrane lipids. The analysis of the 9 superfamilies revealed amino acid distribution patterns in agreement with their known functions and membrane-binding mechanisms. Besides revealing novel amino acids patterns at protein-membrane interfaces, our work contributes a new PMP dataset and an analysis pipeline that can be further built upon for future studies of PMPs properties, or for developing PMPs prediction tools using for example, machine learning approaches.

SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

MOTIVATION: Many bioinformatics areas require us to assign domain matches onto stretches of a query protein. Starting with a set of candidate matches, we want to identify the optimal subset that has limited/no overlap between matches. This may be further complicated by discontinuous domains in the input data. Existing tools are increasingly facing very large data-sets for which they require prohibitive amounts of CPU-time and memory. RESULTS: We present cath-resolve-hits (CRH), a new tool that uses a dynamic-programming algorithm implemented in open-source C++ to handle large datasets quickly (up to ∼1 million hits/second) and in reasonable amounts of memory. It accepts multiple input formats and provides its output in plain text, JSON or graphical HTML. We describe a benchmark against an existing algorithm, which shows CRH delivers very similar or slightly improved results and very much improved CPU/memory performance on large datasets. AVAILABILITY AND IMPLEMENTATION: CRH is available at https://github.com/UCLOrengoGroup/cath-tools; documentation is available at http://cath-tools.readthedocs.io. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Bioinformatics (2015) 31(21):3460–3467. Author, Sayoni Das, would like to report a typographical error in Equation (2) in the above article.

The MSAViewer is a quick and easy visualization and analysis JavaScript component for Multiple Sequence Alignment data of any size. Core features include interactive navigation through the alignment, application of popular color schemes, sorting, selecting and filtering. The MSAViewer is 'web ready': written entirely in JavaScript, compatible with modern web browsers and does not require any specialized software. The MSAViewer is part of the BioJS collection of components. AVAILABILITY AND IMPLEMENTATION: The MSAViewer is released as open source software under the Boost Software License 1.0. Documentation, source code and the viewer are available at http://msa.biojs.net/Supplementary information: Supplementary data are available at Bioinformatics online. CONTACT: [email protected].

MOTIVATION: Computational approaches that can predict protein functions are essential to bridge the widening function annotation gap especially since <1.0% of all proteins in UniProtKB have been experimentally characterized. We present a domain-based method for protein function classification and prediction of functional sites that exploits functional sub-classification of CATH superfamilies. The superfamilies are sub-classified into functional families (FunFams) using a hierarchical clustering algorithm supervised by a new classification method, FunFHMMer. RESULTS: FunFHMMer generates more functionally coherent groupings of protein sequences than other domain-based protein classifications. This has been validated using known functional information. The conserved positions predicted by the FunFams are also found to be enriched in known functional residues. Moreover, the functional annotations provided by the FunFams are found to be more precise than other domain-based resources. FunFHMMer currently identifies 110,439 FunFams in 2735 superfamilies which can be used to functionally annotate>16 million domain sequences. AVAILABILITY AND IMPLEMENTATION: All FunFam annotation data are made available through the CATH webpages (http://www.cathdb.info). The FunFHMMer webserver (http://www.cathdb.info/search/by_funfhmmer) allows users to submit query sequences for assignment to a CATH FunFam. CONTACT: [email protected] SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

In order to understand the evolution of enzyme reactions and to gain an overview of biological catalysis we have combined sequence and structural data to generate phylogenetic trees in an analysis of 276 structurally defined enzyme superfamilies, and used these to study how enzyme functions have evolved. We describe in detail the analysis of two superfamilies to illustrate different paradigms of enzyme evolution. Gathering together data from all the superfamilies supports and develops the observation that they have all evolved to act on a diverse set of substrates, whilst the evolution of new chemistry is much less common. Despite that, by bringing together so much data, we can provide a comprehensive overview of the most common and rare types of changes in function. Our analysis demonstrates on a larger scale than previously studied, that modifications in overall chemistry still occur, with all possible changes at the primary level of the Enzyme Commission (E.C.) classification observed to a greater or lesser extent. The phylogenetic trees map out the evolutionary route taken within a superfamily, as well as all the possible changes within a superfamily. This has been used to generate a matrix of observed exchanges from one enzyme function to another, revealing the scale and nature of enzyme evolution and that some types of exchanges between and within E.C. classes are more prevalent than others. Surprisingly a large proportion (71%) of all known enzyme functions are performed by this relatively small set of 276 superfamilies. This reinforces the hypothesis that relatively few ancient enzymatic domain superfamilies were progenitors for most of the chemistry required for life.

Predicting protein function from structure remains an active area of interest, particularly for the structural genomics initiatives where a substantial number of structures are initially solved with little or no functional characterisation. Although global structure comparison methods can be used to transfer functional annotations, the relationship between fold and function is complex, particularly in functionally diverse superfamilies that have evolved through different secondary structure embellishments to a common structural core. The majority of prediction algorithms employ local templates built on known or predicted functional residues. Here, we present a novel method (FLORA) that automatically generates structural motifs associated with different functional sub-families (FSGs) within functionally diverse domain superfamilies. Templates are created purely on the basis of their specificity for a given FSG, and the method makes no prior prediction of functional sites, nor assumes specific physico-chemical properties of residues. FLORA is able to accurately discriminate between homologous domains with different functions and substantially outperforms (a 2-3 fold increase in coverage at low error rates) popular structure comparison methods and a leading function prediction method. We benchmark FLORA on a large data set of enzyme superfamilies from all three major protein classes (alpha, beta, alphabeta) and demonstrate the functional relevance of the motifs it identifies. We also provide novel predictions of enzymatic activity for a large number of structures solved by the Protein Structure Initiative. Overall, we show that FLORA is able to effectively detect functionally similar protein domain structures by purely using patterns of structural conservation of all residues.

At UCL we have developed several automated protocols for generating protein family resources (CATH; Gene3D). These resources can be used to perform comparative genome analyses in order to understand the evolution of protein families. Also to identify biologically and/or medically interesting families for which no structural data currently exists and which may therefore be important targets for structure genomics initiatives.The CATH domain structure database, established by Orengo and Thornton in 1993, now contains a significant proportion of protein structures from the PDB clustered into 1400 evolutionary families. Relationships have been identified using robust structure comparison methods (SSAP, CATHEDRAL). We have also benchmarked and optimised various 1D-profiles and HMM based protocols for assigning genome sequences to families within the resource (e.g. SAM-T99, SAMOSA, CATH-ISL).In this way we can assign structural data to a large proportion (up to 60%) of whole or partial sequences in completed genomes and >80% of genes coding for enzymes and other proteins in biochemical pathways. However, in order to include all families regardless of whether their structure is known or not, a new protein family resource has been developed (Gene3D). In Gene3D, complete genes have been clustered according to sequence similarity alone, using a robust clustering method (Pfscape). 120 completed genomes from all kingdoms have been clustered into 220,000 gene families, 70,000 of which contain 2 or more sequences. Subsequently, we have labelled those gene families for which CATH structural or Pfam functional domain annotations can be provided for all or part of the gene.Preliminary analysis of the genome annotations reveals that a significant proportion (up to 70%) of CATH annotated genes or gene regions in genomes are assigned to domain families that are common to all three kingdoms of life. However, only 20% of the genome sequences are assigned to gene families common to all kingdoms. Since a large proportion of these genes are multidomain proteins this supports the view that a great deal of functional diversity within the genomes has been achieved by combining domain modules in different ways.In collaboration with Professor Janet Thornton, we have analysed a subset of 56 bacterial genomes to determine the recurrence of specific domain structure families within the genomes. This revealed a small but essential group of universal, and in some cases, highly recurring domain families. For some size-dependent families, domain recurrence is highly correlated with increase in genome size, whilst in other size-independent families no correlation is observed. Statistical analysis allowed us to distinguish three groups. Within the size-dependent families we differentiated two groups: linearly-distributed and non-linearly-distributed. Functional annotation using the COGs revealed that these domains were predominantly involved in metabolism and regulation, respectively. Whilst a third group of Evenly-distributed size independent domains are primarily involved in protein translation and biosynthesis.By mapping CATH and Pfam domains families onto all the genome sequences in Gene3D we observe that a few hundred highly recurrent families are dominating at least 50% of whole or partial genome sequences. Many of these families are common to both prokaryotes and eukaryotes and are performing essential generic functions. In many of the largest families, significant divergence in sequence has been accompanied by modifications in structure and function. Targetting representatives in these families for structure determination will allow the structure genomics initiatives to map both fold and function space and reveal the mechanisms by which divergence in protein families promotes evolution of new functions.

This paper reports a graph-theoretic program, GRATH, that rapidly, and accurately, matches a novel structure against a library of domain structures to find the most similar ones. GRATH generates distributions of scores by comparing the novel domain against the different types of folds that have been classified previously in the CATH database of structural domains. GRATH uses a measure of similarity that details the geometric information, number of secondary structures and number of residues within secondary structures, that any two protein structures share. Although GRATH builds on well established approaches for secondary structure comparison, a novel scoring scheme has been introduced to allow ranking of any matches identified by the algorithm. More importantly, we have benchmarked the algorithm using a large dataset of 1702 non-redundant structures from the CATH database which have already been classified into fold groups, with manual validation. This has facilitated introduction of further constraints, optimization of parameters and identification of reliable thresholds for fold identification. Following these benchmarking trials, the correct fold can be identified with the top score with a frequency of 90%. It is identified within the ten most likely assignments with a frequency of 98%. GRATH has been implemented to use via a server (http://www.biochem.ucl.ac.uk/cgi-bin/cath/Grath.pl). GRATH's speed and accuracy means that it can be used as a reliable front-end filter for the more accurate, but computationally expensive, residue based structure comparison algorithm SSAP, currently used to classify domain structures in the CATH database. With an increasing number of structures being solved by the structural genomics initiatives, the GRATH server also provides an essential resource for determining whether newly determined structures are related to any known structures from which functional properties may be inferred.