Universal Features in the Genome-level Evolution of Protein Domains

TL;DR

This paper introduces a stochastic model based on Chinese Restaurant Processes that explains how protein domain distributions and family counts depend universally on genome size, highlighting a decreasing innovation rate with larger genomes.

Contribution

The authors develop a simple, universal model that accounts for genome size-dependent trends in protein domain distributions and family counts using only two parameters.

Findings

Model reproduces power-law distributions of domain classes

Number of domain families is sublinear in genome size

Innovation probability decreases with genome size

Abstract

Protein domains are found on genomes with notable statistical distributions, which bear a high degree of similarity. Previous work has shown how these distributions can be accounted for by simple models, where the main ingredients are probabilities of duplication, innovation, and loss of domains. However, no one so far has addressed the issue that these distributions follow definite trends depending on protein-coding genome size only. We present a stochastic duplication/innovation model, falling in the class of so-called Chinese Restaurant Processes, able to explain this feature of the data. Using only two universal parameters, related to a minimal number of domains and to the relative weight of innovation to duplication, the model reproduces two important aspects: (a) the populations of domain classes (the sets, related to homology classes, containing realizations of the same domain in different proteins) follow common power-laws whose cutoff is dictated by genome size, and (b) the number of domain families is universal and markedly sublinear in genome size. An important ingredient of the model is that the innovation probability decreases with genome size. We propose the possibility to interpret this as a global constraint given by the cost of expanding an increasingly complex interactome.