MODEL LEARNING PATH
SpliceAI
SpliceAI predicts donor and acceptor probabilities along a DNA sequence. Its variant workflow aligns both alleles, averages five networks and reports the largest gain or loss near the variant.
Loading the interactive page…
A landmark in long-context splice prediction with a concrete downstream record: the 2023 ClinGen SVI Splicing Subgroup calibrated SpliceAI scores for computational evidence within variant interpretation. That is not a claim that a score alone diagnoses disease.
Chapters
Prepare transcript context
Sequence flanks and gene orientation define the input window.
Orient DNA windows consistently before comparing alleles.
Predict splice sites
Dilated residual convolutions use the flanks as context; cropping and a pointwise head produce central-region probabilities.
Use flanking sequence context while predicting only the retained central positions.
Combine aligned predictions
Average the five networks, restore genomic orientation and align reference/alternate positions across indels.
Align ensemble and allele predictions before comparing splice-site scores.
Summarize changes or train
Gain/loss scores summarize allele differences; categorical supervision trains the site predictor.
Summarize donor/acceptor gain or loss from aligned predictions.
All exercises in this guide
- Reverse-complement sequences and align strand-aware outputs
- Edit and crop an alternate-allele DNA window
- Crop, orient and average genomic predictions
- ReLU
- Dense 1-D dilated convolution
- Masked Sum
- Masked Mean
- Layer Normalization
- BatchNorm training, evaluation and functional running statistics
- Affine Projection
- Activations and gated feed-forward transitions
- Affine projection and embedding lookup
- Align insertion/deletion predictions to reference coordinates
- Splice Variant Effects
- Splice and chromatin variant-effect scores
- Masked Log-Sum-Exp
- Categorical Cross-Entropy
- Masked Softmax
- Stable categorical and binary classification losses