This code runs the analyses presenting in the paper, which has been published in PNAS and is available here. In order to run any of the code, you must install the dependencies via
$ conda env create -f environment.yml
which will create an anaconda environment named fitness_env.
All plots presented in the main text and SI of the paper are made by running the scripts in the scripts folder.
Many of these scripts load pre-calculated results that are stored in the results folder.
There are two major analyses that one may wish to re-run, rather than loading the pre-calculated results. First, LASSO estimates of
the empirical and quasi-empirical fitness functons can be re-calculated using the scripts/run_lasso_exp.py script, which is run as:
$ python run_lasso_exp.py <data_name>
where <data_name> can be either tagbfp, his3p_small, his3p_big, or rna.
Next, the numerical tests to calculate a suitable value of the scripts/run_C_calculation.py script. Please
run
$ python run_C_calculation.py --help
for information on the command line arguments required to run the script.
Although it is not used in any of our analyses, the function sample_gnk_fitness_function in src/gnk_model.py may be of interest to some.
This function samples a fitness function from the GNK model given the sequence length L, alphabet size q, and a set of neighborhoods V.
V can be set to one of the string values "random", "adjacent" or "block" if one wishes to use one of the standard neighborhood schemes; in this case,
the optional K parameter must be also be input. Otherwise, V is a list of 1-indexed lists where the i-th list corresponds to the neighborhood
of position i. An example where we sample a L=3, q=2 fitness function with custom neighborhoods is below:
V = [[1, 2], [2], [1, 2, 3]]
f = sample_gnk_fitness_function(3, 2, V=V)