This repository contains the scripts and programs to reproduce all the results from our paper:

Haplotype-based inference of the distribution of fitness effects

Diego Ortega-Del Vecchyo, Kirk E. Lohmueller and John Novembre

Programs and example pipeline

We included the programs PReFerSim and FoIS in the Programs/ folder along with an example pipeline in the folder Programs/ExamplePipeline . There is a README document in the Programs/ExamplePipeline folder with more instructions on how to run our method to estimate the DFE using the pairwise haplotypic identity by state lengths L:

a) PReFerSim.- We conducted forward simulations using PReFerSim. We include the version of PReFerSim we used in our paper with this repository. Check https://github.com/LohmuellerLab/PReFerSim for the latest version of PReFerSim and the instructions to run PReFerSim.

b) FoIS.- This program implements an importance sampling method based on a paper by Monty Slatkin (2001, Genetics Research) to simulate a set of allele frequency trajectories from genetic variants evolving under a particular strength of natural selection. We performed some modifications to the original method by Monty Slatkin to:

1. Model the uncertainty in our estimate of the population allele frequency.
2. Evaluate multiple values of selection using the same set of simulated allele frequency trajectories.
   

We use this program to calculate the expected value of statistics from alleles that have a particular sample allele frequency in the present, such as the pairwise identity-by state lengths surrounding variants that have a particular strength of natural selection acting on them. The program can model any arbitrary demographic scenario. The options of FoIS can be found in README_FoIS.txt. The files Programs/ExamplePipeline/README.md and Programs/ExamplePipeline/SimulateUsingISRoutine.sh include an example of how to run FoIS.

c) ExamplePipeline.- This pipeline includes scripts that can be used to

1. Generate the pairwise identity by state lengths L.
2. Generate the table that computes the likelihoods of L(4Ns, allele frequency, Demographic scenario | L) for a single selection coefficient 4Ns (see equation 2 from our paper).
3. Generate a table that computes the likelihoods of L(alpha, beta, allele frequency, Demographic scenario | L) for two parameters alpha and beta of a partially collapsed gamma distribution of DFEf, the distribution of fitness effects of variants at a certain frequency (also see equation 3 from our paper).
4. Compute the maximum likelihood estimate of either a) the single selection coefficient 4Ns or b) the two parameters alpha and beta of the DFEf.
5. Estimate the distribution of fitness effects of new variants, DFE, from the DFEf.
6. Calculate L from data.
7. An ABC algorithm to estimate demographic history based on the L values at neutral sites.

Plotting scripts

The folder PlottingScripts/ includes the R scripts we used to create each figure from the paper.

Commands to reproduce the analysis associated to main figures from our paper

Check README_ForwardSims.txt for instructions to run forward simulations with PReFerSim that lead to the results shown in figures 2, 4 and 8.

Check README_FoIS.txt for instructions to infer the distribution of fitness effects of variants at a particular frequency. This leads to the results shown in figures 3, 5 and 6.

Check README_DFEUK10K.txt for instructions to infer the distribution of fitness effects of new variants in the UK10K dataset. These analysis lead to the results shown in Figure 9. The location of the L values calculated from the data, as well as the instructions to generate them, is detailed on README_DFEUK10K.txt .

Check README_SI_Figs.txt for instructions to create the data for the supplementary figures.