What indices can we use to describe fitness landscapes? We usually talk of smooth or rugged fitness landscape. * Are there any (standard) indices to measure the "structure" of fitness landscapes? * For example, one might consider the mean epistatic interactions (over all possible combination of loci)

This paper uses 4 metrics for discreet landscapes.

<

Deviation from additivity: How much your genes interact (normalized). Fitness in a purely additive landscape with three genes, for example, is F=f(x)+g(y)+h(z) where f,g, and h are each functions of one variable (the x,y, and z "genes"/coordinates). Deviation measures how far from additivity the fitness is. Higher is not always rougher but it is "trickier" since the effects of changing a gene depend on other genes.

Peak fraction: The % of points that are local peaks (discreet landscapes only), higher is rougher.

Tree component: The % of points that have at most one neighbor of higher fitness. Lower is rougher and indicates that there is a choice as to which direction to travel.

% monotonic paths: The fraction of paths (of all _possible shortest_ paths from a given point to the peak) that don't go downhill. This calculation is averaged over all starting points in the landscape (excluding the peak). Lower is rougher.