The conserved domain database in 2023

Abstract

INTRODUCTION AND OVERVIEW

CDD aims to collect a comprehensive set of protein and domain family models, and it does allow for considerable redundancy in the model set, to ensure good coverage of the protein space. Models that provide significantly overlapping annotation are clustered into protein domain superfamilies, and when domain annotation fails to exceed critical model-specific score thresholds, CDD by default reports superfamily annotation rather than individual model hits. For each model, we compute a consensus sequence, which is used for display purposes only, and reflects the length of the position-specific score matrix (PSSM). While consensus sequences are visible and made available, CDD is not a sequence collection, but is rather meant to enrich the annotation of existing sequence collections. The current CDD version, v3.20, contains 59 693 protein- and protein domain-models obtained from Pfam (1) version 34, SMART (2), the COGs collection (3), TIGRFAMS (4), the NCBI Protein Clusters collection (5), NCBIfam (6) and CDD’s in-house data curation effort (7) (about 1600 new or updated models). For CDD v3.20, the fixed assumed size of the domain model database has again been increased to match the current size of the model collection, resulting in marginally higher E-values reported by RPS-BLAST (8). The upcoming CDD release v3.21 will contain a revised COGs collection, Pfam version 35 and around 500 new or updated models from CDD’s in-house curation. Table ​Table11 details the composition of CDD release v3.20 and the contributions from each source database.

The sizes of these source databases vary considerably, as they were assembled for distinct purposes, and may only consider subsets of the protein space, such as limited by taxonomy. Neither of them is a complete accounting of protein and protein domain families, and neither is the aggregate of many resources, as (1) available protein sequence collections are expanding rapidly and becoming more diverse and (2) limited curation resources impact further growth.

Table ​Table22 shows the 20 largest classifications for common and functionally diverse domain families that have recently been updated or added to CDD. Hierarchical classifications are revisited as curation resources permit. Quite commonly, the analysis of novel ‘dark matter’ families suggests membership in a previously established superfamily classification, for example, or the availability of newly determined 3D structures suggests changes to the multiple sequence alignments that may have an impact on classification details. Over 4700 models curated by the CDD group have been newly published or updated since CDD release v3.16. As of now, a total of 42 937 site annotations are available on 15 943 out of 18 882 CDD staff-curated domain models. Sequence patterns have been recorded for 4971 of these site annotations, so that pattern matching can be used to decide whether annotation can be mapped on to individual sequences.

We identify as ‘dark matter’ those protein sequences in a collection of non-redundant representatives from a taxonomically diverse set, which do not yet have conserved domain annotation via CDD. These are clustered into sequence-similar groups, and multiple sequence alignment models are created for clusters that contain either (1) sequences obtained from experimentally determined 3D structure, or (2) sequences associated with publications, unless these publications describe very large sequence sets such as complete genomes. These alignment models together with the sources that may provide functional information are triaged by curation staff and selected for inclusion in CDD if generic functional information can be provided, as a minimum.

SPARCLE superfamily architectures

We define protein domain architectures as the sequential (N- to C-terminal) list of one or more domain footprints annotated on a protein sequence. SPARCLE, which stands for ‘Subfamily Protein Architecture Labeling Engine’, groups proteins by domain architecture, and now considers both specific and superfamily domain architectures. Superfamily architectures collect proteins that have significant hits to one or more domains that do not exceed a model-specific bitscore thresholds that labels them as high-confidence or ‘specific’. Typically, they will be assigned more generic names, while specific domain architectures may group more closely releated proteins and be assigned more precise and informative names and functional descriptions. To date, CDD curators have reviewed and assigned names and functional labels to about 40,000 well-represented architectures, with a focus on those common in bacterial genomes and those providing coverage for the genomes of eukaryotes such as fungi, protists, and nematodes, in work aimed at supporting the NIH Comparative Genome Resource (CGR, https://www.ncbi.nlm.nih.gov/comparative-genomics-resource/). A publicly accessible Entrez database supports text queries into protein family models that include not only architectures, but also HMMs, and NCBI BLAST rules, and points to summary information as well as links to other databases and sources of attribution.

For the past year, we added attribution to conserved domain architecture, which can be mapped to individual architecture member sequences. Attribution sources are (i) citations, recorded as PubMed IDs, (ii) E.C. numbers (9), (iii) GO terms (10), (iv) gene symbols, (v) TCDB identifiers (11), (vi) CAZy identifiers (12), (vii) MEROPS identifiers (13), (viii) SCOP (14) and/or CATH (15) identifiers. These are assigned and validated by curation staff and are now displayed on CD-Search and BATCH CD-Search pages if the user query matches the corresponding curated domain architecture (see Figure ​Figure11).

Open in a separate window

Figure 1.

BATCH CD-Search results formatted for a few query sequences demonstrate the availability of domain architecture information (under the heading ‘Protein classification’), as well as transferable attributes assigned to each architecture, on top of domain footprint annotation and functional sites associated with some of the domain models. The protein classification information and site annotations can be toggled off for a sparse display focusing on domain footprints only.

SPARCLE architecture curation supports the automated, evidence-based assignment of names to proteins in RefSeq and the Prokaryotic Genome Annotation Pipeline (PGAP) (6). At this time, about 60 million bacterial RefSeq proteins are named via SPARCLE (out of 195 million total bacterial proteins and 160 million proteins with naming evidence provided).

DATA AVAILABILITY

Table ​Table33 lists URLs for major services, tools, and data collections provided by CDD. RPS-BLAST is a part of NCBI’s BLAST software distribution, while CDD provides pre-formatted RPS-BLAST search databases, so that conserved domain searches can be run locally. The rpsbproc stand-alone utility uses additional data to format RPS-BLAST results so that they agree with reports generated by CD-Search (16) and BATCH CD-Search, including domain superfamily assignments and site annotations. The sparclbl (SparcleLabel) utility processes results from local RPS-BLAST searches and provides suggestions for protein names based on domain architecture and is used in NCBI’s prokaryotic genome annotation pipeline (PGAP) (6).

CONCLUSION

The major focus of ongoing work at CDD is to ensure sustainability of the data flow and public services. We strive to keep up with the growth of the sequence collections and to provide a source of annotation that maintains relatively high coverage while improving accuracy. We continue to add novel domain family models to the collection and to establish hierarchical classifications of selected protein domain families where they will have a significant impact on protein naming by domain architecture. We are investigating whether we’ll be able to provide meaningful filtering of search results by taxonomy, for cases where the taxonomic source of the user query sequence is known, and how this knowledge could be used to speed up searching.

ACKNOWLEDGEMENTS

Contributor Information

Jiyao Wang, National Center for Biotechnology Information, National Library of Medicine, National Institutes of Health, Bldg. 38 A, Room 8N805, 8600 Rockville Pike, Bethesda, MD 20894, USA.

Farideh Chitsaz, National Center for Biotechnology Information, National Library of Medicine, National Institutes of Health, Bldg. 38 A, Room 8N805, 8600 Rockville Pike, Bethesda, MD 20894, USA.

Myra K Derbyshire, National Center for Biotechnology Information, National Library of Medicine, National Institutes of Health, Bldg. 38 A, Room 8N805, 8600 Rockville Pike, Bethesda, MD 20894, USA.

Noreen R Gonzales, National Center for Biotechnology Information, National Library of Medicine, National Institutes of Health, Bldg. 38 A, Room 8N805, 8600 Rockville Pike, Bethesda, MD 20894, USA.

Marc Gwadz, National Center for Biotechnology Information, National Library of Medicine, National Institutes of Health, Bldg. 38 A, Room 8N805, 8600 Rockville Pike, Bethesda, MD 20894, USA.

Shennan Lu, National Center for Biotechnology Information, National Library of Medicine, National Institutes of Health, Bldg. 38 A, Room 8N805, 8600 Rockville Pike, Bethesda, MD 20894, USA.

Gabriele H Marchler, National Center for Biotechnology Information, National Library of Medicine, National Institutes of Health, Bldg. 38 A, Room 8N805, 8600 Rockville Pike, Bethesda, MD 20894, USA.

James S Song, National Center for Biotechnology Information, National Library of Medicine, National Institutes of Health, Bldg. 38 A, Room 8N805, 8600 Rockville Pike, Bethesda, MD 20894, USA.

Narmada Thanki, National Center for Biotechnology Information, National Library of Medicine, National Institutes of Health, Bldg. 38 A, Room 8N805, 8600 Rockville Pike, Bethesda, MD 20894, USA.

Roxanne A Yamashita, National Center for Biotechnology Information, National Library of Medicine, National Institutes of Health, Bldg. 38 A, Room 8N805, 8600 Rockville Pike, Bethesda, MD 20894, USA.

Mingzhang Yang, National Center for Biotechnology Information, National Library of Medicine, National Institutes of Health, Bldg. 38 A, Room 8N805, 8600 Rockville Pike, Bethesda, MD 20894, USA.

Dachuan Zhang, National Center for Biotechnology Information, National Library of Medicine, National Institutes of Health, Bldg. 38 A, Room 8N805, 8600 Rockville Pike, Bethesda, MD 20894, USA.

Chanjuan Zheng, National Center for Biotechnology Information, National Library of Medicine, National Institutes of Health, Bldg. 38 A, Room 8N805, 8600 Rockville Pike, Bethesda, MD 20894, USA.

Christopher J Lanczycki, National Center for Biotechnology Information, National Library of Medicine, National Institutes of Health, Bldg. 38 A, Room 8N805, 8600 Rockville Pike, Bethesda, MD 20894, USA.

Aron Marchler-Bauer, National Center for Biotechnology Information, National Library of Medicine, National Institutes of Health, Bldg. 38 A, Room 8N805, 8600 Rockville Pike, Bethesda, MD 20894, USA.

REFERENCES