isweep 1.0.0

Methods for selective sweep inference

See announcements.md for high-level updates on this repo.

See usage.md to evaluate if this methodology fits your study.

Citation

Please cite if you use this package.

Temple, S.D., Waples, R.K., Browning, S.R. (2023) "Modeling recent positive selection in Americans of European ancestry" https://www.biorxiv.org/content/10.1101/2023.11.13.566947v1

Methodology

Acronym: Incomplete Selective sweep With Extended haplotypes Estimation Procedure

This software presents statistical methods to study very recent, very strong positive selection.

• By very recent, we mean within the last 500 generations
• By very strong, we mean selection coefficient s >= 0.015 (1.5%)

The methods relate the lengths of IBD tracts to a coalescent model under selection.

We assume 1 selected allele at a biallelic locus.

There are ~ seven major ideas.

1. A genome-wide selection scan that looks for anomalously large IBD rates
2. Inferring anomalously large IBD clusters
3. Ranking alleles based on evidence for selection
4. Computing a measure of cluster agglomeration (an IBD information entropy)
5. Estimating allele frequency of unknown sweeping allele
6. Estimating selection coefficients (w/ nice statistical properties)
7. Estimating a confidence interval (w/ nice statistical properties)

These steps are implemented automatically in a snakemake pipeline.

Installation

1. git clone https://github.com/sdtemple/isweep.git
2. conda env create -f isweep-environment.yml

• conda activate isweep
• python -c 'import site; print(site.getsitepackages())'

3. Download software

• You need to cite these software
• bash get-software.sh software
  • Puts these in a folder called software/
  • Requires wget
• For simulation study, download SLiM yourself
  • Put in software/
  • https://messerlab.org/slim/

Use IBDkin (https://github.com/YingZhou001/IBDkin) to remove close relatives.

Use PCA or ADMIXTURE to subset based on ancestry.

If you want to compare against other methods (using our pipelines), you require more software.

See workflow/other-methods/ folder.

Overview

This repository contains a Python package and some Snakemake bioinformatics pipelines.

• The package ---> src/
• The pipelines ---> workflow/

You should run all snakemake pipelines in their workflow/some-pipeline/.

You should be in the conda activate isweep environment for analyses.

You should run the analyses using cluster jobs.

We have made README.md files in most subfolders.

The input data is:

• Whole genome sequences
  • Probably at least > 500 diploids
  • Phased vcf data 0|1
  • No apparent population structure
  • No apparent close relatedness
  • A genetic map (bp ---> cM)
  • Recombining diploid chromosomes
    • Not extended to human X chromosome (yet?)
• Access to cluster computing
  • You should have at least 25 Gb of RAM and 6 CPUs on a node
    • More for larger datasets
  • Have not extended to cloud computing (yet?)

Running the procedure:

This is the overall procedure. You will see more details for each step in workflow/some-pipeline/README.md files.

Phase data w/ Beagle or Shapeit beforehand.

1. Make pointers to large (phased) vcf files
2. Edit yaml files in the different workflow directories
3. Run the selection scan (workflow/scan)

• nohup snakemake -s Snakefile-scan.smk -c1 --cluster "[options]" --jobs X --configfile *.yaml &
• Recommendation: Do a test run with your 2 smallest chromosomes.
• Check the *.log files from ibd-ends. If it recommends an estimated err, change the error rate in yaml.
• Then, run with all your chromosomes.

4. Estimate recent effective sizes (workflow/scan)

• workflow/scan/scripts/run-ibdne.sh

5. Make the Manhattan plot (workflow/scan)

• workflow/scan/scripts/manhattan.py

6. Checkout the roi.tsv file

• cat roi.tsv
• Edit it with locus names if you want

7. Run the region of interest analysis (workflow/roi)

• nohup snakemake -s Snakefile-roi.smk -c1 --cluster "[options]" --jobs X --configfile *.yaml &

Tip: define variables for file, folder names, e.g., variable1=1224 then echo $variable1

Considerations

• Positive selection
  • s >= 0 for estimation (use bounds in scipy.optimize.minimize_scalar)

Development things to do

• Provide option for model selection, standing variation, etc. in roi.tsv
• Provide some scripts to summarize analyses
• Test performance in array data, less dense sequence data
• Test performance in dominance selection (sequence data)
• Time varying selection coefficients?
• Not designed for ploidy != 2 (yet)

Current bugs