Boost f-statitics power using genealogies. Compatible with admixtools2.
Full documentation at leospeidel.github.io/twigstats.
Each function is described on this page: leospeidel.github.io/twigstats/reference.
Please see zenodo.org/records/13880459 for scripts to run Relate and Twigstats on your data.
Installation
Please install this package by running the following command in R:
library(remotes)
install_github("leospeidel/twigstats")or
install.packages('twigstats', repos = c('https://leospeidel.r-universe.dev', 'https://cloud.r-project.org'))Alternatively, clone this directory (https://github.com/leospeidel/twigstats) and then in R type
If you have issues installing the package
Please make sure to have an up-to-date R version (>=3.6.0) and C/C++ compiler (e.g., >=GCC v8) loaded in your environment. If you encounter issues, it can help to create a clean R library. Assuming you place this in your home directory, this is done using
mkdir ~/R_libs_for_twigstats/
export R_LIBS_USER="~/R_libs_for_twigstats/"Now, all R packages will be installed into this fresh R library. So please proceed by installing the package as described above.
Basic Usage
- Please see leospeidel.github.io/twigstats/articles/basic-usage.html.
- For a small real data example see leospeidel.github.io/twigstats/articles/real-data-example.html.
- We also provide a detailed example, including scripts that can be repurposed for other datasets, at zenodo.org/records/13880459. We demonstrate how to prepare the data, run Relate, run Twigstats, infer admixture proportions, and plot an MDS.
Simulations
We simulate five populations named PI, PO, P1, P2, and PX, as described in our paper (Speidel et al, 2024). In this simulation PX is admixed between P1 and P2 50 generations ago and these source groups split 250 generations ago.
The proportion is constrained by a ratio of two f4-statistics f4(PO, PI, PX, P1)/f4(PO, PI, P2, P1). We compute these f-statistics on the genotypes, as conventional, and using Twigstats choosing a cutoff time of 500 generations. We additional compare against a version where we compute f-statistics on variants ascertained using a rare-variant cutoff of 5% MAF, and we compare against an idealised ‘chromosome painting’ approach.
We find that admixture proportions are inferred accurately and without bias using Twigstats.
Real data
We provide a detailed example data set, including ancient genomes from Britain, Ireland, and Scandinavia. We plot an MDS and compute admixture proportions in ‘Anglo-Saxon’ indiviuals in Britain using an f4-ratio statistic. For more details, please see zenodo.org/records/13880459.