rfitness: Generate random fitness.

Description Usage Arguments Details Value Author(s) References See Also Examples

View source: R/rfitness.R


Generate random fitness landscapes under a House of Cards, Rough Mount Fuji, or additive model.


rfitness(g, c = 0.5, sd = 1, mu = 1, reference = "random", scale = NULL,
         wt_is_1 = c("subtract", "divide", "force", "no"),
         log = FALSE, min_accessible_genotypes = NULL,
         accessible_th = 0, truncate_at_0 = TRUE)



Number of genes.


The decrease in fitness of a genotype per each unit increase in Hamming distance from the reference genotype (see reference).


The standard deviation of the random component (a normal distribution of mean mu and standard deviation sd).


The mean of the random component (a normal distribution of mean mu and standard deviation sd).


The reference genotype: for the deterministic, additive part, this is the genotype with maximal fitness, and all other genotypes decrease their fitness by c for every unit of Hamming distance from this reference. If "random" a genotype will be randomly chosen as the reference. If "max" the genotype with all positions mutated will be chosen as the reference. If you pass a vector (e.g., reference = c(1, 0, 1, 0)) that will be the reference genotype. If "random2" a genotype will be randomly chosen as the reference. In contrast to "random", however, not all genotypes have the same probability of being chosen; here, what is equal is the probability that the reference genotype has 1, 2, ..., g, mutations (and, once a number mutations is chosen, all genotypes with that number of mutations have equal probability of being the reference).


Either NULL (nothing is done) or a two-element vector. If a two-element vector, fitness is re-scaled between scale[1] (the minimum) and scale[2] (the maximum).


If "divide" (the default) the fitness of all genotypes is divided by the fitness of the wildtype (after possibly adding a value to ensure no negative fitness) so that the wildtype (the genotype with no mutations) has fitness 1. This is a case of scaling, and it is applied after scale, so if you specify both "wt_is_1 = 'divide'" and use an argument for scale it is most likely that the final fitness will not respect the limits in scale.

If "subtract" we shift all the fitness values (subtracting fitness of the wildtype and adding 1) so that the wildtype ends up with a fitness of 1. This is also applied after scale, so if you specify both "wt_is_1 = 'subtract'" and use an argument for scale it is most likely that the final fitness will not respect the limits in scale (though the distorsion might be simpler to see as just a shift up or down).

If "force" we simply set the fitness of the wildtype to 1, without any divisions. This means that the scale argument would work (but it is up to you to make sure that the range of the scale argument includes 1 or you might not get what you want). Note that using this option can easily lead to landscapes with no accessible genotypes (unless you also use scale).

If "none", the fitness of the wildtype is not touched.


If TRUE, log-transform fitness.


If not NULL, the minimum number of accessible genotypes in the fitness landscape. A genotype is considered accessible if you can reach if from the wildtype by going through at least one path where all changes in fitness are larger or equal to accessible_th. The changes in fitness are considered at each mutational step, i.e., at each addition of one mutation we compute the difference between the genotype with k + 1 mutations minus the ancestor genotype with k mutations. Thus, a genotype is considered accessible if there is at least one path where fitness increases at each mutational step by at least accessible_th.

If the condition is not satisfied, we continue generating random fitness landscapes with the specified parameters until the condition is satisfied.

(Why check against NULL and not against zero? Because this allows you to count accessible genotypes even if you do not want to ensure a minimum number of accessible genotypes.)


The threshold for the minimal change in fitness at each mutation step (i.e., between successive genotypes) that allows a genotype to be regarded as accessible. This only applies if min_accessible_genotypes is larger than 0. So if you want to allow small decreases in fitness in successive steps, use a small negative value for accessible_th.


If TRUE (the default) any fitness <= 0 is substituted by a small positive constant (1e-9). Why? Because MAGELLAN and some plotting routines can have trouble (specially if you log) with values <=0. Or we might have trouble if we want to log the fitness.


The model used here follows the Rough Mount Fuji model in Szendro et al., 2013 or Franke et al., 2011. Fitness is given as

f(i) = -c d(i, reference) + x_i

where d(i, j) is the Hamming distance between genotypes i and j (the number of positions that differ) and x_i is a random variable (in this case, a normal deviate of mean mu and standard deviation sd).

Setting c = 0 we obtain a House of Cards model. Setting sd = 0 fitness is given by the distance from the reference and if the reference is the genotype with all positions mutated, then we have a fully additive model (fitness increases linearly with the number of positions mutated).

For OncoSimulR, we often want the wildtype to have a mean of 1. Reasonable settings are mu = 1 and wt_is_1 = 'subtract' so that we simulate from a distribution centered in 1, and we make sure afterwards (via a simple shift) that the wildtype is actuall 1. The sd controls the standard deviation, with the usual working and meaning as in a normal distribution, unless c is different from zero. In this case, with c large, the range of the data can be large, specially if g (the number of genes) is large.


An matrix with g + 1 columns. Each column corresponds to a gene, except the last one that corresponds to fitness. 1/0 in a gene column denotes gene mutated/not-mutated. (For ease of use in other functions, this matrix has class "genotype_fitness_matrix".)

If you have specified min_accessible_genotypes > 0, the return object has added attributes accessible_genotypes and accessible_th that show the number of accessible genotypes under the specified threshold.


Ramon Diaz-Uriarte


Szendro I.~G. et al. (2013). Quantitative analyses of empirical fitness landscapes. Journal of Statistical Mehcanics: Theory and Experiment\/, 01, P01005.

Franke, J. et al. (2011). Evolutionary accessibility of mutational pathways. PLoS Computational Biology\/, 7(8), 1–9.

See Also

oncoSimulIndiv, plot.genotype_fitness_matrix, evalAllGenotypes allFitnessEffects plotFitnessLandscape


## Random fitness for four genes-genotypes,
## plotting and simulating an oncogenetic trajectory

r1 <- rfitness(4)
oncoSimulIndiv(allFitnessEffects(genotFitness = r1))

Bioconductor-mirror/OncoSimulR documentation built on May 31, 2017, 9:37 p.m.