Configuration
SpliceAI2 two-model multispecies ensemble (Jaganathan et al. 2026)
SpliceAI2 two-model multispecies ensemble: 196,608 bp sense-strand input with species one-hot; about 13 million trainable parameters per model; donor, acceptor and junction usage; shared candidate splice sites for both alleles and both models.
Overview
An explanatory profile has not yet been reviewed for this record.
Consult the linked sources for architecture or protocol details. Missing evidence is not evidence of a missing capability.
Evaluations and results
10 evaluations · 23 results. Different protocols are not a single leaderboard.
Sorted by Spearman (all variants; essential splice variants excluded) (higher is better). The best value in each column is highlighted. Decimals are rounded for display; each value links to the printed value and its source.
| Tested configuration | Protocol and dataset | Spearman (all variants; essential splice variants excluded) | Spearman (exonic variants; essential splice variants excluded) | Spearman (intronic variants; essential splice variants excluded) | AUPRC (all variants; essential splice variants excluded) | AUPRC (all variants) | AUPRC (deep intronic variants (500 bp or more from the nearest exon); essential splice variants excluded) | AUPRC (exonic variants; essential splice variants excluded) | AUPRC (near intronic variants (under 500 bp from the nearest exon); essential splice variants excluded) | AUROC (all variants; essential splice variants excluded) | AUROC (exonic variants; essential splice variants excluded) | AUROC (intronic variants; essential splice variants excluded) | Count (excess panel-matched variants over phenotype-shuffled expectation at panel-match odds ratio 2) | Pearson (with mean carrier plasma protein z-score; more negative is better) | Spearman (deep intronic variants (500 bp or more from the nearest exon); essential splice variants excluded) | Spearman (near intronic variants (under 500 bp from the nearest exon); essential splice variants excluded) | Spearman (with multi-cohort raw allele count in high-pLI genes; more negative is better) | True positives (of the 23 prioritised variants, those showing RNA-seq evidence of the predicted splicing change) | Count (rare variants prioritised at SpliceAI2 > 0.25 in panel genes) |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| SpliceAI2 two-model multispecies ensemble (Jaganathan et al. 2026)Evaluation | GTEx v8 fine-mapped sQTL effect size prediction (SpliceAI2 paper, Spearman), GTEx v8 sQTL fine-mapped and null SNVs near junction ends (Jaganathan et al. 2026) | 0.73 (best in column) | 0.7 (best in column) | 0.8 (best in column) | Not reported | Not reported | Not reported | Not reported | Not reported | Not reported | Not reported | Not reported | Not reported | Not reported | Not reported | Not reported | Not reported | Not reported | Not reported |
| SpliceAI2 two-model multispecies ensemble (Jaganathan et al. 2026)Evaluation | Cryptic splice site usage among RNA-seq-confirmed GT/AG-creating SNVs, GTEx (SpliceAI2 paper, Spearman), GTEx rare GT/AG-creating SNVs with RNA-seq cryptic splicing labels (Jaganathan et al. 2026) | 0.63 | 0.66 | Not reported | Not reported | Not reported | Not reported | Not reported | Not reported | Not reported | Not reported | Not reported | Not reported | Not reported | 0.54 | 0.67 | Not reported | Not reported | Not reported |
| SpliceAI2 two-model multispecies ensemble (Jaganathan et al. 2026)Evaluation | OpenSplice MPRA delta PSI prediction for 283,121 single-nucleotide substitutions in 599 exons (SpliceAI2 paper, Spearman), OpenSplice MPRA single-nucleotide substitutions with measured delta PSI (as used by Jaganathan et al. 2026) | 0.59 | 0.62 | 0.52 | Not reported | Not reported | Not reported | Not reported | Not reported | Not reported | Not reported | Not reported | Not reported | Not reported | Not reported | Not reported | Not reported | Not reported | Not reported |
| SpliceAI2 two-model multispecies ensemble (Jaganathan et al. 2026)Evaluation | Cryptic splice site activation by rare GT/AG-creating variants in Genomics England blood RNA-seq (SpliceAI2 paper, auPRC), Genomics England rare-disease participants with WGS and blood RNA-seq (Jaganathan et al. 2026) | Not reported | Not reported | Not reported | Not reported | 0.59 | Not reported | Not reported | Not reported | Not reported | Not reported | Not reported | Not reported | Not reported | Not reported | Not reported | Not reported | Not reported | Not reported |
| SpliceAI2 two-model multispecies ensemble (Jaganathan et al. 2026)Evaluation | Excess panel-matched rare variants in Genomics England rare-disease probands at panel-match odds ratio 2 (SpliceAI2 paper), Genomics England 100,000 Genomes Project probands with dominant-disorder panels (Jaganathan et al. 2026) | Not reported | Not reported | Not reported | Not reported | Not reported | Not reported | Not reported | Not reported | Not reported | Not reported | Not reported | 133 | Not reported | Not reported | Not reported | Not reported | Not reported | Not reported |
| SpliceAI2 two-model multispecies ensemble (Jaganathan et al. 2026)Evaluation | RNA-seq confirmation of SpliceAI2-prioritised rare variants in panel genes, Genomics England rare-disease participants (SpliceAI2 paper), Genomics England rare-disease participants with WGS and blood RNA-seq (Jaganathan et al. 2026) | Not reported | Not reported | Not reported | Not reported | Not reported | Not reported | Not reported | Not reported | Not reported | Not reported | Not reported | Not reported | Not reported | Not reported | Not reported | Not reported | 17 | 23 |
| SpliceAI2 two-model multispecies ensemble (Jaganathan et al. 2026)Evaluation | Cryptic splice site activation by rare GT/AG-creating SNVs, GTEx WGS and RNA-seq (SpliceAI2 paper, auPRC), GTEx rare GT/AG-creating SNVs with RNA-seq cryptic splicing labels (Jaganathan et al. 2026) | Not reported | Not reported | Not reported | 0.77 | Not reported | 0.68 | 0.8 | 0.79 | Not reported | Not reported | Not reported | Not reported | Not reported | Not reported | Not reported | Not reported | Not reported | Not reported |
| SpliceAI2 two-model multispecies ensemble (Jaganathan et al. 2026)Evaluation | GTEx v8 fine-mapped sQTL SNVs against non-prioritised SNVs near junction ends (SpliceAI2 paper, auROC), GTEx v8 sQTL fine-mapped and null SNVs near junction ends (Jaganathan et al. 2026) | Not reported | Not reported | Not reported | Not reported | Not reported | Not reported | Not reported | Not reported | 0.76 | 0.73 | 0.8 | Not reported | Not reported | Not reported | Not reported | Not reported | Not reported | Not reported |
| SpliceAI2 two-model multispecies ensemble (Jaganathan et al. 2026)Evaluation | Depletion of high-scoring splice-proximal SNVs in gnomAD, TOPMed and UK Biobank, high-pLI genes (SpliceAI2 paper, Spearman), gnomAD, TOPMed and UK Biobank splice-proximal SNV depletion set (Jaganathan et al. 2026) | Not reported | Not reported | Not reported | Not reported | Not reported | Not reported | Not reported | Not reported | Not reported | Not reported | Not reported | Not reported | Not reported | Not reported | Not reported | -0.122 | Not reported | Not reported |
| SpliceAI2 two-model multispecies ensemble (Jaganathan et al. 2026)Evaluation | Carrier plasma protein levels for rare splice-region variants, UK Biobank (SpliceAI2 paper, Pearson), UK Biobank rare splice-region variants with carrier plasma protein levels (Jaganathan et al. 2026) | Not reported | Not reported | Not reported | Not reported | Not reported | Not reported | Not reported | Not reported | Not reported | Not reported | Not reported | Not reported | -0.504 | Not reported | Not reported | Not reported | Not reported | Not reported |
All 23 result rows with coverage, uncertainty and sources
Filter evaluations
Applied filters: All linked evaluations
Source checking is not independent reproduction. Release 2026-10-22-1b7cf027e4f6.
Use this configuration
How it works, versions and access
Strengths, limitations and unresolved questions
Evidence
Source checking verifies the cited claim or transcription. It does not establish independent reproduction.
Evidence table
Inspect claims, sources and review details
Trace each statement to its source and review. A context-only reference supports the record generally; it does not verify an individual field. Source checking does not reproduce an experiment.
One row per statement and cited source. Multiple citations are not independent evaluations. Shared locators are labelled explicitly.
0 evidence rows matching the loaded filters
| Property and statement | Original source and location | Review and provenance |
|---|
No evidence rows match these filters. Choose another scope or clear the search.
Sources and history
Release 2026-10-22-1b7cf027e4f6 · Record review: source checked
3 source records and release history
- A unified framework for quantitative splicing and transcript prediction (Jaganathan et al., bioRxiv 2026) · Original source · bioRxiv preprint version 1, posted 2026-10-11; full-text PDF (57 pages)
- SpliceAI2 README at commit 1dcd2a1 (Illumina/SpliceAI2) · Original source · Commit 1dcd2a1882a8223095e7529982ae70bcb658340a (main, 2026-10-11); two README-only commits after tag v2.0 (aa634374bc8700414b1a5388729faed7c4f7ba07)
- SpliceAI2 Model Terms of Use (LICENSE at commit 1dcd2a1, Illumina/SpliceAI2) · Original source · Commit 1dcd2a1882a8223095e7529982ae70bcb658340a; file unchanged since tag v2.0
Technical metadata and extraction receipts
Stable ID: rna-splicing-20261013-spliceai2-config-spliceai2-ensemble
- areas
- dna-genomes
- contexts
- research
- method types
- supervised_machine_learning
- foundation model eligible
- false
- reported name
- SpliceAI2
- version
- Two models trained with identical settings and averaged (bioRxiv v1 Methods); code release v2.0
- parameters
- 196,608 bp sense-strand input with species one-hot; about 13 million trainable parameters per model; donor, acceptor and junction usage; shared candidate splice sites for both alleles and both models; summary score is the maximum of donor gain, donor loss, acceptor gain and acceptor loss deltas
- training overlap
- Trained on all human genes except those on chromosomes 21 and 22 (held out for validation). The variant benchmarks are genome-wide, so most benchmark variants lie in genes seen in training. Supplementary Table 1 lists GTEx among the recount3 human training sources; the paper does not say whether junctions observed in the GTEx benchmark samples were removed from training.
- access reuse
- Code at Illumina/SpliceAI2 under the SpliceAI2 Model Terms of Use (non-commercial organisations only; research use only). Weights on Hugging Face illumina-ai/SpliceAI2, gated.
- code commit
- 1dcd2a1882a8223095e7529982ae70bcb658340a
- source locator
- Methods 'SpliceAI2 model' (Training, Variant effect prediction); README; LICENSE
- missing metadata
- checkpoint: reason: unreported; note: The paper does not say whether the evaluated ensemble is the released Hugging Face checkpoints (spliceai2_multispecies_13M_01 and _02).
- provenance note
- SpliceAI2 two-model multispecies ensemble (Jaganathan et al. 2026), as evaluated in the SpliceAI2 paper.
Related records
- configuration of: SpliceAI2
- system: SpliceAI2 on Genomics England GT/AG cryptic splice activation (auPRC, SpliceAI2 paper)
- system: SpliceAI2 excess panel-matched variants at odds ratio 2 (Genomics England)
- system: SpliceAI2 prioritised variants confirmed in Genomics England RNA-seq
- system: SpliceAI2 on GTEx GT/AG (auprc, SpliceAI2 paper)
- system: SpliceAI2 on GTEx GT/AG (spearman-correlation, SpliceAI2 paper)
- system: SpliceAI2 on GTEx sQTL (auroc, SpliceAI2 paper)
- system: SpliceAI2 on GTEx sQTL (spearman-correlation, SpliceAI2 paper)
- system: SpliceAI2 on OpenSplice MPRA (spearman-correlation, SpliceAI2 paper)
- system: SpliceAI2 on splice-proximal SNV depletion in high-pLI genes (Spearman, SpliceAI2 paper)
- system: SpliceAI2 on UK Biobank carrier protein levels (Pearson, SpliceAI2 paper)