DPCfam

DPCfam performs unsupervised clustering of homologous protein regions from sequence alignments using Density Peak Clustering to classify protein families and support generation of functional and structural hypotheses for entries in UniProt and Pfam.

Key Features:

• Unsupervised classification: Operates without labeled data to classify homologous protein regions based on sequence alignments.
• Density Peak Clustering (DPC): Employs Density Peak Clustering to detect cluster centers and assign sequences to clusters.
• Sequence alignment mapping: Uses sequence alignments to map homologous regions within proteins.
• Pfam correspondence and discovery: Generates clusters that often correspond to one or more Pfam families and reveals unassigned sections of sequence space.
• Homology-based hypothesis generation: Identifies homologous relationships to facilitate in silico generation of functional and structural hypotheses.
• Domain discovery and classification improvement: Facilitates domain discovery, detection of classification inconsistencies, improvement of family coverage, and enhancement of clan membership.
• Environmental metagenomics applicability: Applicable to unsupervised classification of sparsely annotated protein datasets from environmental metagenomics and analysis of domain diversity.

Scientific Applications:

• Manual annotation assistance: Aids domain discovery, detection of classification inconsistencies, improvement of family coverage, and enhancement of clan membership within databases such as Pfam.
• Environmental metagenomics: Enables unsupervised classification and domain diversity analysis of sparsely annotated protein datasets derived from environmental metagenomics.
• Hypothesis generation for UniProt entries: Supports generation of functional and structural hypotheses for proteins in UniProt that lack experimental validation.

Methodology:

Maps homologous protein regions using sequence alignments and applies Density Peak Clustering (DPC) to detect cluster centers and assign sequences to clusters; clusters are compared to Pfam families and a proof-of-principle analysis was performed on two Pfam clans.