On the Degree of Freedom in Multilevel Evolutionary Models

Biotic systems have a high degree of freedom subject to evolution, but also have constraints and semi-invariant structure (e.g. RNA/protein folding is not random). A common approach to model biological evolution is to construct a " biologically " pre-structured system with a few mutable parameters—i.e. a small degree of freedom—and to study the strengthening/weakening of the prede-fined structure (example below). Although we consider it essential for better understanding, this approach does not allow us to study the emergence of novel structures. To go beyond this, we recently investigated a multilevel evolutionary model that incorporated both a high degree of freedom and biologically relevant structure (Takeuchi and Hogeweg 2008). We first review this study and then compare it with another study where we investigated an evolutionary model with a considerably smaller degree of freedom (Takeuchi and Hogeweg 2009); therewith we present some interesting insight on the the degree of freedom in evolutionary models. We investigated the evolution of RNA-like replicator systems by explicitly modeling the genotype-phenotype-interaction mapping of individuals (Takeuchi and Hogeweg 2008). By using RNA folding and base-pair-matching based molecular recognition, we incorporated complex structure and a high degree of freedom into the genotype-phenotype-interaction mapping. The results showed that a population, originally consisting of one genotype, evolved into a complex ecosystem consisting of up to four quasi-species through a chain reaction of niche generation and speciation, where a replicase species generated a niche for a parasite species, of which evolution generated a niche for an " escaping " replicase species and so on. Through this diversification, evolution generated unique classes of genotypes and phenotypes having distinct ecological functionality, which was not " built-in " to the model. That said, let us now consider more of a behind-the-scenes story. Initially, we were confronted with the results of the simulations, which were just millions of genotypes present in the system over time, an apparently unintelligible mess of data. Among numerous means to analyze the data, bioinformatic pattern detection devices (viz. phylogenetic trees and sequence-logo) turned out to be the most useful in this case. These devices enabled us to realize the existence of sequence classes and associated sequence patterns; stated differently, we discovered " meaningful observables " of the system. We reran simulations with each individual being identified by observables " designed " to distinguish the recognized sequence classes. The results revealed a sequence of " events described in