How do reference genome choice and pipeline versions affect variant calls?

Why

A read is aligned relative to a specific reference assembly. If the reference sequence changes, reads near gaps, duplicated regions or structural differences may map differently, which affects apparent alleles and genotype likelihoods. Genome coordinates and even the reference allele can differ between assemblies, so old and new VCF positions cannot safely be compared as plain numbers.

Software versions and resource bundles also matter. A new caller, filtering model, truth set or annotation database may improve detection while changing the list of reported calls. To make results reproducible, pin the assembly and reference checksum, sequence dictionary, aligner and caller versions, parameters, quality filters, reference datasets and workflow container or commit.

When changing pipelines, reprocess validation samples and compare sensitivity, precision and difficult genomic regions against appropriate benchmarks. Coordinate liftover is useful but not a substitute for validating biological equivalence.