A Network Filtration Protocol for Elucidating Relationships between Families in a Protein Similarity Network

TL;DR

This paper introduces a new network filtration protocol that enhances the analysis of protein similarity networks, enabling clearer delineation of family boundaries and relationships within enzyme superfamilies, thereby improving functional predictions.

Contribution

The authors developed a novel algorithm that filters similarity networks to identify family boundaries and relationships, providing a more quantitative and interpretable view of protein superfamily topology.

Findings

The protocol accurately estimates family boundaries within protein similarity networks.

Neighboring families in the kinase superfamily tend to share structural and functional properties.

Network topology correlates with phylogenetic data, validating the method's effectiveness.

Abstract

Motivation: The study of diverse enzyme superfamilies can provide important insight into the relationships between protein sequence, structure and function. It is often challenging, however, to discover these relationships across a large and diverse superfamily. Contemporary similarity network visualization techniques allow researchers to aggregate sequence similarity information into a single global view. Network visualization provides a qualitative estimate of functional diversity within a superfamily, but is unable to quantitate explicit boundaries, when present, between neighboring families in sequence space. This limits the potential of existing sequence-based algorithms to generate functional predictions from superfamily datasets. Results: By building on current network analysis tools, we have developed a new algorithm for elucidating pairs of homologous families within a sequence dataset. Our algorithm is able to filter through a dense similarity network in order to estimate both the boundaries of individual families and also how the families neighbor one another. Globally, these neighboring families define a topology across the entire superfamily. The topology is simple to interpret by visualizing the network output generated by our filtration protocol. We have compared the network topology within the kinase superfamily against available phylogenetic data. Our results suggest that neighbors within the filtered kinase network are more likely to share structural and functional properties than more distant network clusters.