The CATH database is a free, publicly available online resource that provides information on the evolutionary relationships of protein domains. It was created in the mid-1990s by Professor Christine Orengo and colleagues, and continues to be developed by the Orengo group at University College London.

Experimentally-determined protein three-dimensional structures are obtained from the Protein Data Bank and split into their consecutive polypeptide chains, where applicable. Protein domains are identified within these chains using a mixture of automatic methods and manual curation. The domains are then classified within the CATH structural hierarchy: at the Class (C) level, domains are assigned according to their secondary structure content, i.e. all alpha, all beta, a mixture of alpha and beta, or little secondary structure; at the Architecture (A) level, information on the secondary structure arrangement in three-dimensional space is used for assignment; at the Topology/fold (T) level, information on how the secondary structure elements are connected and arranged is used; assignments are made to the Homologous superfamily (H) level if there is good evidence that the domains are related by evolution, i.e. they are homologous. To browse the classification hierarchy, see CATH hierarchy.

Additional sequence data for domains with no experimentally determined structures are provided by our sister resource, Gene3D, which are used to populate the homologous superfamilies. Protein sequences from UniProtKB and Ensembl are scanned against CATH HMMs to predict domain sequence boundaries and make homologous superfamily assignments.

CATH has been recognized as a Global Core BioData Resource (GCBR) by the Global Biodata Consortium. This endorsement reflects the database's significance as a reliable and comprehensive resource for protein structure classification in the life sciences community.

We are pleased to announce the release of CATH v4.4 (October 2024 ; https://www.cathdb.info/), the latest update to the CATH (Class, Architecture, Topology, Homology) structural classification database. This release includes an up-to-date classification of PDB structures as well as over 90 million domain models from The Encyclopedia of Domains (TED) (https://ted.cathdb.info).

CATH v4.4 incorporates approximately ~600.000 newly classified domain structures from the Protein Data Bank (PDB) and maps over 90 million predicted domain structures from the Encyclopedia of Domains (TED) resource into CATH superfamilies—a joint effort between the Jones group (UCL Computer Science) and the Orengo group (UCL Structural and Molecular Biology). This integration has resulted in a 180-fold increase in structural information for CATH superfamilies.

The inclusion of TED data has expanded the number of superfamilies from 5,841 to 6,573, folds from 1,349 to 2,081, and architectures from 41 to 77. It is important to note that the TED data comprises predicted structures, and these new folds and architectures remain hypothetical until experimentally confirmed.

Advancements in Domain Segmentation and Classification: To manage the substantial volume of data from AlphaFold Protein Structure Database, our automated domain segmentation workflow has been enhanced. We have integrated a faster and more accurate in-house deep-learning approach called Chainsaw, along with the publicly available methods Merizo and UniDoc. For homologue detection and verification, we predicted CATH superfamilies using a deep-learning tool based on embeddings from protein language models, CATHe, and expanded our suite of protein structure comparison tools to include Foldseek. Domains from PDB and TED with a homology assignment were further validated using Hidden Markov Model matching with strict overlaps and manual curation. These advancements enable the classification of domains into CATH superfamilies using evidence from multiple independent approaches, including both structural and sequence-based methods.

Expansion of Functional Families (FunFams): Within superfamilies, CATH further subclassifies domains into coherent sets of sequences where functions are conserved, called Functional Families (FunFams). We updated the sequences in FunFams to UniProt release 2024_02, achieving a 276% increase in FunFam coverage. Additionally, the mapping of TED structural domains has resulted in a fourfold increase in FunFams with structural information, increasing the number of FunFams with at least one high-quality structural representative to 73,215. This expansion enhances our ability to analyze conserved residues within protein families and to identify putative functional sites, contributing to a deeper understanding of protein function and evolution.

Identification of Novel Folds and Architectures: Analysis of TED data has led to the identification of 479 new folds and 34 new architectures, including structures such as the Alpha-propeller, Beta hairpin barrel, and Alpha-Beta flower. These new categories are currently hypothetical and await experimental confirmation.

Future Directions: The extensive data integrated into CATH v4.4 presents opportunities for further exploration of protein structures and evolutionary relationships. Ongoing efforts will focus on refining algorithms and workflows to improve domain boundary assignments, particularly in complex structures such as repeats and proteins with large interfaces.

We provide a daily snapshot of the very latest classifications and annotations as they happen in our pipeline. This enables users to find the most up-to-date information about their particular structure of interest. The amount of data we provide at this stage is limited mainly to domain boundaries and superfamily classification.

We aim to provide full releases of CATH (CATH-Plus) every 12 months. CATH-Plus adds a significant amount of data on top of the core classification information available in CATH. The CATH-Plus release process includes a number of manual annotation checks in addition to adding a huge amount of information combining protein structure, sequence and function. As a result, there is a greater depth of information available in CATH-Plus, though it may be missing information on the most recent structures.

CATH-Plus data includes:

The homologous superfamilies in CATH-Gene3D can often be functionally and structurally diverse even though they share a conserved structural core. Therefore, the superfamilies have been sub-classified into functional families (FunFams) using a subclassification protocol purely based on sequence patterns. Relatives within these FunFams are likely to share highly similar structures and functions. The FunFams are useful in function prediction and in providing information on the evolution of function.

The structures within a homologous superfamily have been clustered at < 9 Å RMSD to form structural clusters, also known as structurally-similar groups (SSGs). These structural clusters are useful for understanding the structural diversity of a superfamily.

The conserved structural core in the homologous superfamilies can be observed from the structural superpositions generated from its representative domains by CATH Tools. It is an effective way of observing the structural conservation and diversity across the superfamily.

See release notes for information on the statistics for specific releases.

CATH and CATH-Plus data for all releases can be downloaded from Data Downloads.

CATH is proud to be a member of the open source software community. Our developers use and contribute towards the development and maintenance of a number of open source tools. For a full list of the open source software used in the making of this resource (both in the pipeline and our web pages), please visit the CATH tools page.

If you have any comments/suggestions/criticisms, please let us know:

https://www.cathdb.info/support/contact