Web Application for DPCfam and DPCstruct Data Exploration

The rapid growth of protein sequence and structure databases has created a significant annotation gap that manual curation alone cannot bridge. Two recent computational biology methods, DPCfam and DPCstruct, address this by applying the Density Peak Clustering (DPC) algorithm to automatically group millions of protein domains into putative evolutionary families called metaclusters, based purely on pairwise similarity scores derived from large-scale sequence or structural alignments. This process has yielded 81,384 sequence-based and 28,246 structure-based families with no manual curation. Although the datasets derived from both methods are publicly available on Zenodo under FAIR data principles, exploring them typically requires downloading and parsing gigabyte-scale archives, including a massive 81 GB XML file for DPCfam, which presents a technical barrier for many researchers. In this thesis, we present DPCexplorer, a unified Django web application that closes this gap. Raw archives were preprocessed on the ORFEO HPC cluster and loaded into a PostgreSQL database optimized with B-Tree, GIN trigram, and functional indexes. Built on the Model-View-Template (MVT) pattern, the platform accepts four query types (DPCfam MCID, DPCstruct MCID, Pfam ID, and UniProt accession) and returns paginated metadata tables, per-protein domain architecture diagrams, and downloadable biological files (FASTA, MSA, HMM, PDB). Furthermore, for DPCstruct metaclusters, it integrates the PDBe-Mol* 3D viewer to render AlphaFold2-predicted structures colored by per-residue pLDDT confidence directly in the browser. We further audit the database indexes and assess DPCexplorer against the FAIR Principles for Research Software (FAIR4RS). The application will soon be accessible at https://dpcexplorer.areasciencepark.it and is fully reproducible from the GitHub repository [1] and the associated Zenodo data deposit