ProtoNet

ProtoNet classifies protein sequences into a hierarchical tree to represent protein families across the protein sequence space.

Key Features:

• Unsupervised bottom-up clustering: Uses an unsupervised bottom-up clustering algorithm to organize large protein datasets into a hierarchical tree of clusters.
• All-against-all sequence comparison: Constructs the hierarchical tree based on an all-against-all comparison using 2.5 million representative sequences from UniRef50.
• Database coverage: Includes over 9 million proteins with 5.5% from UniProtKB/SwissProt and the remainder from UniProtKB/TrEMBL.
• Annotation-based quality filtering: Applies rigorous annotation-based tests to identify a refined set of 162,088 high-quality clusters.
• ProtoName assignment: Assigns each cluster a unique ProtoName derived from Gene Ontology (GO) terms, UniProt/Swiss-Prot keywords, and InterPro entries.
• Multi-resolution analysis modes: Provides a default overview of the entire hierarchy and an advanced mode to explore the family tree at varying levels of resolution.
• Version comparison capability: Enables systematic comparison of different ProtoNet versions to observe changes as sequence space knowledge increases.

Scientific Applications:

• Protein family classification: Classifies proteins into hierarchical families to facilitate the study of protein relationships.
• Functional and structural annotation: Supports annotation of clusters via ProtoName assignments using GO, UniProt/Swiss-Prot keywords, and InterPro to identify functional and structural characteristics.
• Large-scale proteome analysis: Enables analysis of proteome-scale data drawn from UniProtKB/SwissProt and UniProtKB/TrEMBL.
• Comparative clustering analysis: Facilitates comparison of clustering outcomes across ProtoNet versions to assess how family definitions change with expanded sequence data.

Methodology:

Hierarchical tree construction via an all-against-all comparison of 2.5 million UniRef50 representative sequences, unsupervised bottom-up clustering to form clusters, annotation-based tests to filter for 162,088 high-quality clusters, and assignment of ProtoNames from GO terms, UniProt/Swiss-Prot keywords, and InterPro entries.