Hierarchy and extremes in selections from pools of randomized proteins
Variation and selection are the core principles of Darwinian evolution, yet quantitatively relating the diversity of a population to its capacity to respond to selection is challenging. Here, we examine this problem at a molecular level in the context of populations of partially randomized proteins selected for binding to well-defined targets. We built several minimal protein libraries, screened them in vitro by phage display and analyzed their response to selection by high-throughput sequencing. A statistical analysis of the results reveals two main findings: first, libraries with same sequence diversity but built around different "frameworks" typically have vastly different responses, second, the distribution of responses within a library follows a simple scaling law. We show how an elementary probabilistic model based on extreme value theory rationalizes these findings. Our results have implications for designing synthetic protein libraries, for estimating the density of functional biomolecules in sequence space, for characterizing diversity in natural populations and for experimentally investigating the concept of evolvability, or potential for future evolution.