0%

​

Share this page with:

Domains tab

The Domains tab provides a domain-level classification of the protein’s structure from The Encyclopedia of Domains (TED).

TED is a computational resource dedicated to the large-scale identification and classification of structural domains within the AlphaFold Database (AFDB). Its methodology employs a state-of-the-art, consensus-based approach that integrates three deep learning-based domain parsing algorithms to robustly determine domain boundaries. 

These domain assignments are then aligned against the CATH database, a widely used structural classification resource, using structural alignment algorithms.

The Domains tab. This view provides detailed information about the structural domains identified by TED, including their CATH classification, boundaries, length, and average pLDDT score.

If there is no domain data available for a particular protein, the Domains tab will display “N/A” and will be disabled. This indicates that a domain classification could not be generated for this entry.

Interactivity between views

The domains track above the PAE, sequence view, and 3D structure viewer are linked to allow interactive exploration of the protein.

Hovering over a domain box above the PAE highlights the same residues in the sequence view and in the 3D structure viewer. This provides a consistent way to locate domains both in the linear sequence and in the folded structure.

Interactive highlighting across views. Hovering the cursor over a specific domain simultaneously highlights the corresponding residues in both the sequence and the 3D viewer.

The 3D structure viewer offers two colouring modes: by domain or by model confidence (pLDDT). Colouring by domain displays distinct structural units, while colouring by model confidence indicates the per-residue confidence score (pLDDT).

These two views can be combined. For example, users may keep the 3D structure coloured by model confidence while hovering over a domain in the track. This makes it possible to simultaneously assess both the structural position of the domain and the prediction reliability of its residues.

A multi-layered view. The protein structure is coloured by model confidence (pLDDT). The boundaries of Domain 1 are simultaneously highlighted in the sequence and 3D viewer.

Qscore

Each domain entry in the AlphaFold Database is associated with a suite of descriptive and quality-assessment metrics. A consensus level serves as an initial quality indicator, reflecting the degree of agreement among the three domain-parsing methods.

For domains with a structural match, a CATH identifier and assignment are provided based on the level of the hierarchy matched. A second key metric, the Qscore, provides a comprehensive, composite measure of the quality of each domain assignment.

The Qscore is a weighted average of several parameters: structural alignment coverage, TED consensus level, domain compactness (globularity), normalised mean pLDDT score, and a modified E-value factor. The formula is as follows:

Quality metric for TED annotations in AlphaFold models. The Qscore is a weighted average of four parameters: structural alignment coverage, consensus level of parsing methods, domain compactness (globularity), and normalised mean pLDDT score. Based on the Qscore and CATH assignment, each domain is given a quality label: “High-confidence” (Qscore ≥ 75 and full 4-level C.A.T.H assignment), “Moderate” (Qscore ≥ 70 and 3-or 4-level C.A.T/C.A.T.H assignment) or “Uncertain” (lack of a 3- or 4-level assignment, regardless of the Qscore).

Based on the Qscore and CATH assignment, each domain is given a quality label:

  • High-confidence: Qscore ≥ 75 and full 4-level C.A.T.H assignment. 
  • Moderate: Qscore  ≥ 70 and 3- or 4-level C.A.T/C.A.T.H assignment.
  • Uncertain:  A lack of a 3- or 4-level assignment, regardless of the Qscore.

This layered approach, in conjunction with residue boundaries, average pLDDT, and Predicted Aligned Error (PAE) from AlphaFold 2 models, provides a powerful framework for understanding how protein domains are organised within the AFDB, offering a valuable resource for evolutionary, structural, and functional studies.