The SUPERFAMILY 2.0 database: a significant proteome update and a new webserver

Abstract Here, we present a major update to the SUPERFAMILY database and the webserver. We describe the addition of new SUPERFAMILY 2.0 profile HMM library containing a total of 27 623 HMMs. The database now includes Superfamily domain annotations for millions of protein sequences taken from the Universal Protein Recourse Knowledgebase (UniProtKB) and the National Center for Biotechnology Information (NCBI). This addition constitutes about 51 and 45 million distinct protein sequences obtained from UniProtKB and NCBI respectively. Currently, the database contains annotations for 63 244 and 102 151 complete genomes taken from UniProtKB and NCBI respectively. The current sequence collection and genome update is the biggest so far in the history of SUPERFAMILY updates. In order to the deal with the massive wealth of information, here we introduce a new SUPERFAMILY 2.0 webserver (http://supfam.org). Currently, the webserver mainly focuses on the search, retrieval and display of Superfamily annotation for the entire sequence and genome collection in the database.


INTRODUCTION
(1) uses a library of 15 438 expertcurated profile hidden Markov models (HMMs) representing protein domains of known structure to predict the presence of structural domains in amino acid sequences. The domain sequences were obtained from the Structural Classification of Protein database (SCOP) (2). SCOP classifies protein domains into Class, Fold, Superfamily and Family level to understand structural, functional and evolutionary relationship between protein structural domains. The Superfamily level domains in SCOP share structural and functional properties that infer common evolutionary origin despite sharing low sequence identity. Whereas at the Family level, most homologous proteins cluster together with high sequence similarity suggesting clear evolutionary relationship and functional consistency (3). The SUPERFAMILY database provides domain annotations at both Superfamily and Family levels (4).
SUPERFAMILY provides various analysis tools to facilitate better analysis and interpretation of the database content. They include the identification of under-and overrepresentation of domains between genomes (5), construction of phylogenetic trees (6), analysis of the domain distribution of superfamilies and families across the tree of life (7) as well as providing ontology based annotations for SUPERFAM-ILY domains and architectures (8,9).
Here we present the development of new SUPERFAM-ILY 2.0 HMM library along with a major database update that includes the addition of SUPERFAMILY annotations for the all the protein sequences from the UniProtKB (10) and NCBI reference genome collections (11). We also introduce a newly developed webserver to mainly focus on the annotation of exponentially growing sequence data as well as to facilitate future integration with the SUPERFAMILY sister resources including dcGO (8) and D 2 P 2 (12) to capture the combined information representing structure, disorder and domain centric ontologies in a single platform. In the following section, we discuss the development of new SUPERFAMILY 2.0 profile HMM library. Later, we discuss the annotation statistics for UniProtKB sequences and NCBI reference genome collection followed by the introduction of the new webserver and its basis functionalities. Finally, we discuss the future directions for the SUPER-FAMILY resource.

SUPERFAMILY 2.0 profile HMM model library
In this update, we have created a new profile HMM library using sequences taken from the structural domain database SCOPe (13), CATH (14), ECOD (15) and PDB (16). Initially, we built the new HMMs for SCOPe domain sequences by filtering it at 95% sequence identity against the * To whom correspondence should be addressed. Tel: +44 122 2267822; Email: apandura@mrc-lmb.cam.ac.uk C The Author(s) 2018. Published by Oxford University Press on behalf of Nucleic Acids Research. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted reuse, distribution, and reproduction in any medium, provided the original work is properly cited.  (Table 1). After the major update, the SUPERFAMILY database contains 50 604 320 and 44 765 365 distinct protein sequences from UniProtKB and NCBI respectively. About 50% of the protein sequences (45 730 297) are common between UniProtKB and NCBI sequence collection. It is worth noting that the annotations for UniProtKB and NCBI sequences were performed using the SUPERFAM-ILY 1.75 HMM library.

New webserver -SUPERFAMILY 2.0
The wealth of proteome sequence information continues to increase manifold with the recent advancement of sequence technologies. In order to meet the challenges involved in the analysis and interpretation of large proteome datasets, we have developed a new webserver (http://supfam.org). The webserver is built using a Perl based real-time web application framework called Mojolicious (https://mojolicious. org). In this new development, we have predominately focused on the search, retrieval and display of Superfamily domain annotations present in the database. We foresee the integration of some of the essential analysis and visualization tools into the new webserver that would eventually replace its predecessor in the near future. In the following section, we discuss some of the key features of the new webserver.
Genome browser. The user can browse all genomes present in the database using the taxonomy tree of life (http:// supfam.org/genome/hierarchy). For easy lookup and navigation, the webpage provides a navigation panel based on the taxonomic class information. Following the links, the user can reach the landing page of a genome that summarizes the annotations statistics (e.g. http://supfam.org/ genome/hs) ( Figure 1A). The genome summary page provides various annotations statistics including the percentage of sequences with Superfamily assignments, percentage of amino acid coverage, number of domains, superfamilies, families, domain pairs and unique architectures that represent predicted combination of Superfamily domains. Following the download links the user can download all protein sequences in fasta format along with its predicted Superfamily assignments as flat text files. The page also provides hyperlinks to access the list of all Superfamily and Family assignments found in the genome which intern pro-vides access to the list of all protein sequences containing the predicted Superfamily and Family domains. A typical domain annotation page for a given protein sequence contains a list of regions on the protein sequence containing the predicted Superfamily and Family domains assignments along with their respective E-values and the closest structure derived from the family assignment procedure (4). The closest structure act as a potential template to build Nucleic Acids Research, 2019, Vol. 47, Database issue D493 comparative models using SUPERFAMILY annotations (20,21). The domain annotation page also contains a graphical representation of the Superfamily domain organisation (shown as coloured blocks) laid on the sequence (shown as a straight line) ( Figure 1B).
Genome statistics. For a quick look up, we provide a summary page listing all genomes present in the database along with the SUPERFAMILY annotations statistics as mentioned above (http://supfam.org/genome/stats). The genomes shown in the statistics page are organized into various model organisms (Eukaryotes, Prokaryotes and their respective strains), metagenome and pseudogenes. Most of the metagenomes annotated in the database were downloaded from the Joint Genome Institute as part of the environmental sequencing project (https://jgi.doe. gov). The pseudogenes are derived computational using inhouse program from the Ensembl genome database (http: //www.ensembl.org/). Various sequence collection including UniProtKB and NCBI are listed in a separate category called 'Others'. Hyperlinks for the genome names provided in this page point directly to the respective genome summary page.
UniProtKB and NCBI genome collection. The majority of protein sequences annotated in the database are from UniprotKB and NCBI resources. These sequences are organised into a set of complete genomes that are available for viewing through the subgenome option (UniPro-tKB genomes: http://supfam.org/subgenome/up, NCBI genomes: http://supfam.org/subgenome/ncb).
Sequence and keyword search. The sequence search facility (http://supfam.org/sequence/search) allows the user to submit up to 1000 protein sequences and obtain its corresponding domain assignments based on the new SUPERF-MAILY 2.0 HMM library. To avoid redundant computation and to speedup, the submitted sequences are searched against the SUPERFAMILY database for exact match with pre-defined domain assignments. Whenever a hit is not found, the sequences are searched against the SCOP domain sequences from ASTRAL (22) using BLAST (23). As a final stage with no hits, the sequences are searched against the SUPERFAMILY 2.0 model library using the HMMER package (17). In addition, the database can be searched using keywords that include protein sequence identifier, genome names, SCOP identifier and SUPER-FAMILY model identifier.

DISCUSSION
We are in the process of integrating the results from the disorder prediction using D 2 P 2 (12) and domain centric ontologies using dcGO (8,9) with SUPERFAMILY domain prediction into a single platform in order to facilitate better interpretation of protein sequence, structure, disorder and function. As part of the Genome3D consortium, SUPER-FAMILY domain predictions are used to build 3D structural models for proteome of various model organisms of significant importance (21). In line with that, the webserver will include new functionality to build 3D models on the fly based in the SUPERFAMILY domain predictions.