GlobalEpistasis

install and load data

• Install julia (see julialang.org, on macos use homebrew).
• Learn some julia (e.g. wikibooks)
• Update packages Pkg.update()
• add package Pkg.clone("https://github.com/jotwin/GlobalEpistasis.jl"), note this will add NLopt, CSV, and DataFrames packages
• I recommend IJulia for an interactive notebook interface.

Load data into julia, CSV doc. Test data is a subset of Wu et al. eLife 2016

using CSV
using DataFrames
dataf = CSV.read("test/wu4sites_test.txt")

prepare data

using GlobalEpistasis
d = prepdata(dataf, :mut, :sequence, "VDGV", :f, cname = :concentration, vname = :v)

arguments are

1. dataframe of your data
2. identifier of the column with sequence data
3. type of sequence data is either :sequence the sequence, or :listofmuts a list of substitutions where each substitution is a string with wild-type amino acid, position, and mutated amino acid, e.g. "V3A". Multiple substitutions per sequence can be concatenated together e.g. "V3A-G2B".
4. wild-type sequence, or if your data is in list of mutants format its the wild-type identifier, e.g. "WT"
5. name of the column with measured phenotype

optional keyword arguments are

• vname: name of the column of variance estimates
• cname: name of the column of conditions, which can be strings or numbers.
• delim: delimiter for multiple substitutions in listofmuts format. Default '-'
• condition_type: whether the condition variable is treated as a continuous or categorical predictor (:categorical or :continuous)

non-epistatic model

@time mlin = nonepistatic_model(d)

output is a dictionary with keys:

• :beta is a data frame of additive effects
• :r2 r-squared
• :sigma2 HOC epistasis variance
• :prediction data frame of measured phenotype (y), inferred measured phenotype (yhat), and inferred additive trait (phi, rescaled), and (optional) conditions
• :rmse root mean squared error
• :ll log-likelihood

global epistasis model

@time m = fit(mlin, nk = 4, tol = 1e-14)

• nk is number of knots in the I-spline (default 4)
• tol is the tolerance of the optimization (it can take a long time or fail if its too small)

output is a dictionary of keys similar to above

writing output

write csv files

CSV.write("betas.txt", m[:beta])
CSV.write("prediction.txt", m[:prediction])

modifying initial conditions

the fit function first argument is an initial model m with entries m[:a] and m[:b]. These can be modified to change the intial parameters. Also you can change the non-epistatic model e.g. mlin[:a] = [0, 0, 0, 0, 0]