google/deepvariant — reverse-engineered prompt

Reverse engineered prompt

GitHub

Build me a genomics command line pipeline for calling DNA variants from aligned sequencing reads. I want users to give it a reference genome, a BAM or CRAM file, an output folder, and choose the data type like whole genome, whole exome, PacBio HiFi, Oxford Nanopore, or hybrid PacBio plus Illumina.

The tool should run the full DeepVariant style flow, turn reads into examples, use a neural network model to call variants, and write standard VCF and gVCF files. Please make it easy to run with Docker, with clear commands, logging, optional VCF stats report, dry run mode, and a setting for number of CPU shards.

Also include sensible handling for human diploid germline calling, optional chrX and chrY haploid settings with PAR regions, and clear warnings about limits like human trained models and non human data. If practical, include DeepTrio support for child and parent samples. Look up current docs online if you need to.

Want more depth? Deep Reverse