In 2014, I did some benchmarking of variant effect prediction algorithms. You can read about that in this blog post.
I used the same input VCF that I created originally. It contains all snps, all 1 base pair insertions and deletions, and 2 possible 2 and 3 base pair insertions and deletions at all locations spanning the CFTR gene with 100bp margins on either side.
You can annotate this vcf with each algorithm using the CWL scripts provided.
docker build --tag="andrewjesaitis/vep" -f Dockerfile.vep . cwltool vep-workflow.cwl cftr-job-vep.yml
Similarly for SnpEff:
docker build --tag="andrewjesaitis/snpeff" -f Dockerfile.snpeff . cwltool snpeff-workflow.cwl cftr-job-snpeff.yml
Then you can open the Jupyter Notebook and rerun all cells.
Otherwise just skip to the punchline and open the notebook on Github.
I've written up a discussion of the results and dug into some particularly troublesome variants on my blog.
I've also added the gzip'd vcf that highligh some mismatches. Note that these vcfs contain repeated variants (since I am outputing a single variant-transcript pair per line). The keys in the
INFO field are self documenting. These files are