Quantifying Deception

Abstract

The concept of in fitness landscapes was introduced in the genetic algorithm (GA) literature to characterize problems where sign epistasis can mislead a GA away from the global optimum. Evolutionary geneticists have long recognized that sign epistasis is the source of the ruggedness of fitness landscapes, and the recent availability of a growing number of empirical fitness landscapes may make it possible for evolutionary biologists to study how deception affects adaptation in a variety of organisms. However, existing definitions of deception are categorical and were developed to characterize landscapes independent of population distributions on the landscape. Here we propose two metrics that quantify deception as continuous functions of the locations of replicators on a given landscape. We develop a discrete population model to simulate within-host evolution on 19 empirical fitness landscapes of Plasmodium falciparum (the most common and deadly form of malaria) under different dosage levels of two anti-malarial drugs. We demonstrate varying levels of deception in malarial evolution, and show that the proposed metrics of deception are predictive of some important aspects of evolutionary dynamics. Our approach can be readily applied to other fitness landscapes and toward an improved understanding of the evolution of antimicrobial drug resistance.