A metacluster (MC) is a group of protein domains that the DPC algorithm has identified as evolutionary relatives: a putative protein domain family. DPCfam metaclusters are built from sequence similarity; DPCstruct metaclusters are built from 3D structural similarity. Each metacluster has a unique ID (e.g., MC1, MC5) and comes with biological properties, seed sequences, and, where available, Pfam annotations.

DPCfam works in sequence space. It clustered roughly 23 million sequences from UniRef50 v. 2017_07 and produced 81,384 metaclusters: a Standard subset (≥ 50 domains, 46,828 metaclusters) and a DPCfamB subset (25 - 49 domains, 34,556 metaclusters).

DPCstruct works in structure space. It clustered roughly 15 million AlphaFold2-predicted structures from the Foldseek Cluster database (a non-redundant representative set of the AlphaFold Database, AlphaFoldDB v4.0) and produced 28,246 structural metaclusters. Because protein structures are more conserved than sequences, DPCstruct is particularly powerful for detecting remote homologs: proteins that share a common fold even when their sequences have diverged well below 30% identity. The average intra-cluster sequence identity in DPCstruct is only 34.5%, placing most families squarely within the twilight zone where standard sequence-based tools become unreliable.

It means that no seed sequence in that metacluster carries a Pfam annotation. These metaclusters are also called "dark matter": protein domain families that exist in nature but have not yet been characterized. They are among the most scientifically interesting results in DPCexplorer: roughly 40% of DPCfam metaclusters (specifically 33,179) and roughly 49% of DPCstruct metaclusters (specifically 13,823) fall into this category, for a total of 47,002 UNKNOWN families. If you have a hypothesis about what one of them might be, we would love to hear from you.

From the home page, select DPCfam or DPCstruct as the database source, type a valid metacluster ID (e.g., MC1 or MC5), and click Search. You will land directly on the detail page for that metacluster, where you can explore its properties, seed sequences, AlphaFold representative structure, and download the associated biological files.

You can also browse all metaclusters by clicking Metaclusters in the top navigation bar and selecting DPCfam or DPCstruct.

Yes. Select Pfam from the landing page, enter a valid Pfam family identifier (e.g., PF02990) or clan identifier (e.g., CL0263), and click Search. The results page will show all DPCfam metaclusters that overlap with that Pfam family (listed by average overlap percentage) and all DPCstruct metaclusters that are annotated with it (listed by Pfam consistency score). Each metacluster ID is a direct link to its detail page.

Yes. Select UniProt from the landing page, enter a valid UniProt accession (e.g., P05067), and click Search. You will see a domain architecture diagram that maps all DPCfam domains, DPCstruct domains, and Pfam annotations onto the full-length protein sequence: three horizontal tracks drawn to scale above and below a ruler line. You can toggle each track on and off using the checkboxes below the diagram.

The 3D viewer is powered by PDBe-Molstar, an open-source molecular viewer developed by the Protein Data Bank in Europe (EMBL-EBI). It runs entirely in your browser using WebGL, so no plugins or external software are required.

For DPCstruct metaclusters, the viewer loads the AlphaFold2-predicted PDB structures of the cluster's representatives directly from our server and colors each residue by its pLDDT confidence score:

For DPCfam metaclusters, the 3D Viewer tab links out to the corresponding entry in the AlphaFold Protein Structure Database, where you can explore the full structural coordinate maps interactively.

From each DPCfam metacluster detail page (Downloads tab):
• Seed sequences (FASTA format)
• Multiple Sequence Alignment (MSA) (FASTA format)
• Profile Hidden Markov Model (HMM)

From each DPCstruct metacluster detail page (Downloads tab):
• Representative seed sequences (FASTA format)
• Representative PDB structures (ZIP archive)

Bulk archives for all metaclusters are also available from the Downloads menu in the navigation bar.

Yes. DPCexplorer now offers a public, read-only REST API under /api/. You can send a single DPCfam or DPCstruct metacluster ID, or a comma-separated list of IDs, and get its properties back as JSON; a separate endpoint returns its member sequences (ID, Protein ID, and coordinate range), paginated. No login or API key is required.

See DPCexplorer_API_Documentation.md on GitHub for the full endpoint reference, including a ready-to-run example Python script and Jupyter notebook. The underlying preprocessed CSV files remain openly available on Zenodo (DOI: 10.5281/zenodo.20159208).

Yes. The full pipeline, from raw Zenodo archives to a running web server, is documented step by step in the repository's README.md. All preprocessing notebooks, Python scripts, and SQL schemas are version-controlled on GitHub under the MIT license. The preprocessed Zenodo deposit also lets you skip the heavy HPC preprocessing and go straight to loading the database. Additionally, full interactive profiling reports for most of the deposited CSV files are available in the repository under static/profiling_reports/ and can be downloaded for interactive exploration in any web browser.

If you use DPCexplorer, please cite the software: 10.5281/zenodo.20575268 (Zenodo concept DOI, always the latest release).

Should you want to learn how DPCexplorer was built, see the MDMC thesis; for a concise summary, see the eScience 2026 paper.

The full bibliography and dataset identifiers are on the About page → References.