Allow 'ANY' txdb, eg. EnsDb objects

[mschubert]

Thank for for the VariantAnnotation package, it makes it very easy to annotate mutations found in VCF files!

There is, however, one limitation that I find unnecessarily restrictive: The predictCoding method expects a TxDb object as its second argument and will not be able to process gene and transcript annotations otherwise.

I propose that the limitation here should be any object that supports the cdsBy and transcriptsBy methods, e.g. an EnsDb object. As they do not share a common base class, I allow ANY type that will then error if the associated methods are not found.

In particular, the code that previously failed now works successfully:

ens106 = AnnotationHub::AnnotationHub()[["AH100643"]]
asm = BSgenome.Hsapiens.NCBI.GRCh38::BSgenome.Hsapiens.NCBI.GRCh38
fname = system.file("extdata", "chr22.vcf.gz", package="VariantAnnotation")
vr = VariantAnnotation::readVcfAsVRanges(fname)
res = predictCoding(vr, ens106, asm)
# Error in (function (classes, fdef, table)  :
#   unable to find an inherited method for 'predictCoding' for signature '"VRanges", "EnsDb", "BSgenome", "missing"'

[Copilot]

Pull request overview

This PR expands predictCoding()’s S4 method dispatch so callers can use transcript annotation providers beyond TxDb (e.g., EnsDb), by relaxing the subject signature to ANY and relying on the presence of the required annotation methods at runtime.

Changes:

• Broaden predictCoding() method signatures from subject="TxDb" to subject="ANY" for IntegerRanges, CollapsedVCF, ExpandedVCF, GRanges, and VRanges entry points.
• Enable predictCoding(vr, ensdb, genome) style calls that previously failed dispatch due to EnsDb not inheriting from TxDb.

Comments suppressed due to low confidence (1)

R/methods-predictCoding.R:53

• The VRanges method calls .predictCoding() directly, so it will bypass any interface validation / cache initialization done in the GRanges method. With subject now ANY, this means VRanges+EnsDb can still fail depending on cdsBy() defaults or missing methods. Consider mirroring the same cache initialization (and clearer error) here before calling .predictCoding().

setMethod("predictCoding", c("VRanges", "ANY", "ANY", "missing"),
    function(query, subject, seqSource, varAllele, ..., ignore.strand=FALSE)
{
    varAllele <- alt(query)
    query <- as(query, "GRanges")

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.