Docker

[rajewski]

Problem

All of the scripts here depend on the UC Riverside HPCC infrastructure in terms of install software like MrBayes, GATK, etc. In addition there are a number of rather off-the-map R packages.

From a reproducibility standpoint, it would be nice to wrap all of this into a single docker container with all dependencies pre-installed.

Solution

Create a Dockerfile to include in the repo that specifies how to build a container to house the analysis. Subsequently, this container should be built and made publicly available in Dockerhub.

Additional considerations

Inside the scripts, file paths will need to be changed. Submission scripts will need to either be removed or reworked to call Docker instead. Removal is probably better since most HPC environments would require Singularity over docker and the submission itself is likely to be very environment-specific (Slurm vs PBS, etc)

[rajewski]

the main problems that I am running into are speedseq and whathap. They both have some complex dependencies which stymy a simple docker container.

[rajewski]

I found a speedseq docker image here, I'm still making a whatshap image, but I've also decided to transition the monolithic container to several smaller ones

• bcftools
• samtools
• bwa
• gatk
• ffq and jq
• speedseq
• whatshap
• R packages

[rajewski]

ugggggh

• plink
• faststructure

Things I've sworn not to include because they would be so onerous to repeat and/or were not used in the final analysis

• RAxML-NG
• MrBayes
• jModelTest
• PArtitionFinder