GRCh38 · PRECOMPUTED VARIANT SCORES
Every scored variant, in context.
Explore splice acceptor and donor gains and losses. Inspect exact alleles, compare model seeds, and see where scoring is available.
Wide summaries combine all ALT alleles and gene annotations. Exact allele, gene and score filters resume when you zoom in.
Drag to pan. Shift-drag marks a region. Click selects an allele. Use + / − to zoom; arrows to pan; R marks the view. Native page scroll and pinch zoom remain available.
Sequence search
N matches A/C/G/T; ambiguous reference bases are not assumed to be known bases.
Model comparison
Matches require the same allele and gene and an unambiguous annotation. REF mismatches are excluded. Seed differences are not calibrated uncertainty or a clinical classification.
Read the tracks
Scores, offsets and coverage
DS_AG and DS_AL describe acceptor gain and loss; DS_DG and DS_DL describe donor gain and loss. Scores range from 0 to 1. A recorded zero is different from an absent prediction. The affected genomic position is variant POS + DP on either gene strand. Offsets attached to zero scores remain available as source values; they do not identify a nonzero predicted event.
At close zoom, the browser fetches exact five-decimal OpenSpliceAI values. Wide views show maxima over annotations in fixed genomic bins. r13 coverage shows accepted source occurrences, rather than predicted variant counts. Unfinished chunks remain visibly pending. Source records can contain repeated alleles, conflicting annotations or a REF mismatch; the inspector preserves those details.
Models and reference
r10 and r13 are individual MANE-trained model seeds with a 10,000 nt flank, ±50 bp event search and M1 masking. The original SpliceAI baseline has two-decimal scores and a different annotation set. Exons and strand come from the scoring annotation; transcript IDs and CDS are not present in that table.
Navigation and exports
Coordinates in the address field and inspector are 1-based inclusive. Internal intervals and CSV start/end columns are 0-based half-open. Shared links capture the immutable dataset, view, model, track order, heights, filters, appearance and selected allele. Figures use the same drawing routine as the visible tracks. Exact CSV preserves all annotations and missing values; summary CSV contains bin statistics and is labelled separately. Exports wait for a successfully loaded, stable view.