Documentation / Reference Databases / SpliceAI Precomputed
SpliceAI precomputed dataset: fields, coverage, and provenance
Folklore reads a managed GRCh38 SpliceAI asset and exposes its donor and acceptor delta-score (DS) and delta-position (DP) fields with source, file-set, and version provenance. Coverage is limited to records represented in that asset. For score meaning and classifier thresholds, use the dedicated predictor guide.
Review one public GRCh38 variant against Folklore's current evidence snapshot. Public search accepts a variant expression only. Do not enter patient or case information.
Search a public variantDatabase Details
Four Delta Scores
SpliceAI produces four model delta scores, each ranging from 0 to 1, representing a predicted change in splice-site usage. The values are not calibrated probabilities that a splice event occurs:
The model predicts increased acceptor usage at a nearby position relative to the reference sequence.
The model predicts reduced usage of an existing acceptor relative to the reference sequence.
The model predicts increased donor usage at a nearby position relative to the reference sequence.
The model predicts reduced usage of an existing donor relative to the reference sequence.
The managed record also exposes the maximum of the four delta scores (max_score). The dedicated predictor guide explains how Folklore interprets that model output; this dataset page documents the asset and fields.
Delta Positions
The paired DP_AG, DP_AL, DP_DG, and DP_DL values report the predicted position of the corresponding splice change relative to the variant. A delta position helps a reviewer locate the predicted event. It does not identify an experimentally observed transcript or determine the functional consequence of that transcript.
Limitations
SpliceAI predictions are based on primary sequence context. Tissue-specific splicing regulation is not modeled.
A managed precomputed dataset has a defined genome build, file set, and coverage. A missing record means the prediction is unavailable in that dataset, not that the variant has no splice effect.
SpliceAI does not establish the transcript-level or functional consequence of a predicted splice change. Its scores are model outputs, not calibrated probabilities of disruption.
The asset does not establish an observed splice effect or its clinical relevance. Score interpretation, RNA evidence, and the full evidence context require qualified review.
References
Jaganathan K, et al. "Predicting splicing from primary sequence with deep learning." Cell. 2019;176(3):535-548. PMID: 30661751.
Walker LC, et al. "Using the ACMG/AMP framework to capture evidence related to predicted and observed impact on splicing." American Journal of Human Genetics. 2023;110(7):1046-1067. PMID: 37352859.
Continue with the SpliceAI predictor guide or review how computational evidence fits the ACMG/AMP framework.