Density Peak Clustering of UniRef50 (DPCfam-UR50)

Density Peak Clustering of UniRef50 (DPCfam-UR50) automates the classification of homologous protein regions from the UniRef50 v.2017_07 sequence database into Metaclusters (MCs) to support annotation and functional prediction of uncharacterized proteins.

Key Features:

• Density Peak Clustering (DPC): Uses Density Peak Clustering to group homologous protein regions into clusters.
• Input dataset: Operates on UniRef50 v.2017_07, containing approximately 23 million sequences.
• Data scale handling: Processes data volumes up to 3 terabytes.
• Metacluster identification: Identifies around 45,000 protein clusters (Metaclusters) representing seed sequence regions.
• Profile HMM support: Metaclusters serve as foundations for building profile Hidden Markov Models (HMMs).
• Cross-resource mapping: Maps clusters to domain resources with approximately 81% of medium-large Pfam families and 72% of ECOD families correspondingly mapped.
• Novel family discovery: Identifies over 14,000 clusters composed of protein regions without Pfam annotation.

Scientific Applications:

• Protein family classification: Classifies unannotated protein sequences into putative families (Metaclusters) to facilitate downstream analyses.
• Annotation and functional prediction: Supports annotation and the generation of testable functional hypotheses for proteins known only at the sequence level.
• Validation against domain databases: Enables cross-referencing of clusters with Pfam and ECOD to validate and contextualize annotations.
• Discovery of novel families: Identifies candidate novel protein families by detecting clusters lacking existing Pfam annotation.

Methodology:

Re-implementation of a previous pipeline optimized for large-scale data that leverages Density Peak Clustering to organize UniRef50 sequences into Metaclusters and facilitates subsequent profile-HMM construction.