AdmixPipe v3: facilitating population structure delimitation from SNP data

Abstract

Summary: Quantifying genetic clusters (=populations) from genotypic data is a fundamental, but non-trivial task for population geneticists that is compounded by: hierarchical population structure, diverse analytical methods, and complex software dependencies. AdmixPipe v3 ameliorates many of these issues in a single bioinformatic pipeline that facilitates all facets of population structure analysis by integrating outputs generated by several popular packages (i.e. CLUMPAK, EvalAdmix). The pipeline interfaces disparate software packages to parse Admixture outputs and conduct EvalAdmix analyses in the context of multimodal population structure results identified by CLUMPAK. We further streamline these tasks by packaging AdmixPipe v3 within a Docker container to create a standardized analytical environment that allows for complex analyses to be replicated by different researchers. This also grants operating system flexibility and mitigates complex software dependencies.

Availability and implementation: Source code, documentation, example files, and usage examples are freely available at https://github.com/stevemussmann/admixturePipeline. Installation is facilitated via Docker container available from https://hub.docker.com/r/mussmann/admixpipe. Usage under Windows operating systems requires the Windows Subsystem for Linux.

Figure 1.

The workflow for AdmixPipe v3 requires two input files: (1) VCF or PLINK-formatted file of genotypes and (2) Tab-delimited population map. Either binary or text format can be used to input PLINK format. Both files are processed by admixturePipeline.py, which handles data filtering steps and executes Admixture. Outputs are then processed through the submitClumpak.py module to drive a local installation of CLUMPAK. Subsequently, distructRerun.py will parse CLUMPAK results and depict outputs. The cvSum.py module calculates CV values and log-likelihood values for all major and minor clusters detected by CLUMPAK. Finally, runEvalAdmix.py will evaluate fit of Admixture models to the genotype data by executing EvalAdmix for all major and minor clusters.

Figure 2.

Comparison of optimal K determination from: (A) K = 8 Admixture-computed CV values, and (B) K = 12 EvalAdmix correlation of residuals. The matrices computed in EvalAdmix (left) show the correlation of residuals for actual and inferred genotypes of all individuals under different Admixture models. Bar plots (right) show the proportion of ancestry for individuals under each Admixture model.