In December 2017, the team at Google Brain released an open source, deep learning based variant caller: DeepVariant (https://github.com/google/deepvariant). DeepVariant outperforms its alternatives by accuracy — it won the accuracy award at the precisionFDA Truth Challenge.
DeepVariant performs variant calling through image recognition. Running it over cloud can get quite expensive or time-consuming. However, turning it into a Nextflow (https://github.com/nextflow-io/nextflow) pipeline allows us to parallelise computations to make use of all cores of the machine and get to the results faster.