Splice variants

Variants near an exon–intron boundary can change how a transcript is spliced. Varcode predicts compatible mechanisms, realizes their transcript products, and compares each with the patient baseline. DNA alone does not establish which transcript is expressed.

Example

This CFTR donor variant uses Ensembl 81 (GRCh38); see reference setup.

from varcode import Variant

variant = Variant("7", 117_531_115, "G", "A", genome=81)
transcript = variant.genome.transcript_by_id("ENST00000003084")
effect = variant.effect_on_transcript(transcript)
print(type(effect).__name__)  # Deletion: the preferred exon-skip comparison
for candidate in effect.candidates:
    print(candidate.effect.short_description)
    print(candidate.effect.mutant_protein_sequence)

No opt-in flag is needed. The top effect is one classified comparison; candidates retains the alternatives, including unresolved products. Read the candidates when evaluating uncertainty; the top description is not an audit of all hypotheses. JSON preserves the full products and evidence.

Candidate outcomes

Mechanisms include normal, exon_skip, intron_retention, and cryptic. The effect class describes the resulting protein change, not the mechanism: exon skipping can produce a Deletion or a frameshift, for example.

for candidate in effect.candidates:
    for outcome in candidate.outcomes:
        print(outcome.hypothesis.baseline_splice_plan)
        print(outcome.hypothesis.splice_plan)
        print(outcome.baseline.cdna_sequence, outcome.mutant.cdna_sequence)

Canonical and exon-skip paths can use transcript annotation. Retention and cryptic paths need the relevant genomic sequence; absent sequence remains None with an unresolved consequence. First and last exons cannot be skipped by this internal-exon mechanism. Plans exclude incompatible combinations such as retaining an intron while skipping an exon it requires.

Coding effects when splicing is unchanged

An exonic variant can alter both a codon and a splice signal. To inspect the conditional coding consequence, select outcomes whose mutant plan uses only normal mechanisms. With germline splice uncertainty, several patient baseline plans may still give different comparisons.

normal_effects = [
    outcome.effect
    for candidate in effect.candidates
    for outcome in candidate.outcomes
    if all(option.mechanism == "normal"
           for _, option in outcome.hypothesis.splice_plan.choices)
]

Ranking and protein sequences

most_likely_candidate.effect is the first ranked comparison; highest_priority_candidate.effect is the most severe consequence. Mechanism preference is ordinal, not a measured probability. Candidate probability remains None when splice choices are uncalibrated. Iterate candidates to retain each sequence's hypotheses and baseline.

Unchanged active splice sites share their mechanism between patient baseline and mutant. If a site changes or disappears, all compatible baseline choices remain. Neither a germline splice change nor an arbitrary preferred mechanism establishes that the patient's normal transcript lacks an exon.

Splice signals

The default uses positional windows, not a learned splice score:

Region	Positions in transcript direction
Donor motif	Intronic +1/+2
Acceptor motif	Intronic −2/−1
Intronic splice region	Donor +3…+6 or acceptor −3
Exonic splice region	Last three exon bases at a donor or first exon base at an acceptor

Removed junction coordinates and changed motif bases produce stronger splice hypotheses than peripheral changes. Unchanged annotated noncanonical sites, including GC–AG and AT–AC introns, remain intact; these are established splice classes (Burset et al.). An altered site still retains normal splicing as a possibility.

RNA evidence

Supplied full RNA models that exactly match a predicted cDNA support every compatible comparison and rank those candidates first. Other alternatives remain. Partial or unmatched models retain their supplied sequence as an unresolved comparison with an unknown baseline. Read counts do not become probabilities; missing reads do not establish normal splicing.

Read collection, alignment and assembly belong upstream in Isovar or another RNA tool. Varcode consumes models via rna_resolver= or a phase resolver; it does not fill missing observed sequence with reference exons. See combining evidence and the RNA evidence API.

Compatibility with Varcode 10

Explicit annotator="fast" retains SpliceOutcomeSet, its mechanism-specific classes, effect_if_splicing_unchanged, and its legacy RNA reconciliation contract. Existing imports and serialized sets remain supported. See the SpliceOutcomeSet reference and migration guide.

Limitations

Variants outside the windows can still affect splicing. Mid-exon regulatory sites, branch points and deep-intronic activation require broader modeling, tracked in #297. Mechanism and protein predictions alone do not prove expression or translation.