Predicting evolution and visualizing high-dimensional fitness landscapes

TL;DR

This paper investigates the structure of high-dimensional fitness landscapes, revealing that peaks are clustered and valleys are shallow, which influences the predictability of evolutionary paths.

Contribution

It introduces methods to visualize and analyze high-dimensional fitness landscapes, highlighting peak clustering and implications for evolutionary predictability.

Findings

High peaks are clustered in fitness landscapes.

Valleys between peaks are shallow and narrow.

Evolutionary trajectories can often reach the highest peak via valley crossings.

Abstract

The tempo and mode of an adaptive process is strongly determined by the structure of the fitness landscape that underlies it. In order to be able to predict evolutionary outcomes (even on the short term), we must know more about the nature of realistic fitness landscapes than we do today. For example, in order to know whether evolution is predominantly taking paths that move upwards in fitness and along neutral ridges, or else entails a significant number of valley crossings, we need to be able to visualize these landscapes: we must determine whether there are peaks in the landscape, where these peaks are located with respect to one another, and whether evolutionary paths can connect them. This is a difficult task because genetic fitness landscapes (as opposed to those based on traits) are high-dimensional, and tools for visualizing such landscapes are lacking. In this contribution, we focus on the predictability of evolution on rugged genetic fitness landscapes, and determine that peaks in such landscapes are highly clustered: high peaks are predominantly close to other high peaks. As a consequence, the valleys separating such peaks are shallow and narrow, such that evolutionary trajectories towards the highest peak in the landscape can be achieved via a series of valley crossings