About
SpliceAI is a state-of-the-art deep learning model developed by Illumina and described in Jaganathan et al., Cell 2019. It is designed to predict the splicing effects of genetic variants — including substitutions, insertions, and deletions — by modeling how changes in DNA sequence affect RNA splice sites. The tool outputs delta scores indicating the likelihood that a variant disrupts or creates donor and acceptor splice sites, enabling researchers and clinicians to prioritize potentially pathogenic variants. Pre-computed annotations for all possible substitutions, 1-base insertions, and 1–4 base deletions within genes are available for download, greatly accelerating variant interpretation workflows. SpliceAI is implemented as a Python package and can be run via command line, making it accessible for integration into bioinformatics pipelines. It supports analysis of both germline and somatic variants and is compatible with standard VCF file formats. The tool has become widely adopted in clinical genetics, rare disease research, and cancer genomics for identifying splice variants that may be missed by conventional annotation tools. Free for academic and not-for-profit use; commercial applications require a license from Illumina.
Key Features
- Deep Learning Splice Prediction: Uses a neural network trained on human genome data to predict the probability that a variant creates or disrupts donor and acceptor splice sites.
- Pre-computed Variant Annotations: Offers downloadable delta score annotations for all possible substitutions, 1-base insertions, and 1–4 base deletions within genes for rapid lookup.
- VCF-Compatible Pipeline Integration: Accepts standard VCF input and outputs annotated VCF files, making it easy to incorporate into existing bioinformatics workflows.
- High Sensitivity for Cryptic Splice Sites: Detects deep intronic and exonic variants that activate cryptic splice sites, which are typically missed by rule-based annotation tools.
- Command-Line Python Package: Distributed as a Python package installable via pip, enabling scriptable, reproducible, and scalable variant annotation on any compute environment.
Use Cases
- Identifying pathogenic splice-altering variants in rare disease patient genomic data
- Prioritizing candidate variants in clinical exome and genome sequencing pipelines
- Detecting cryptic splice site activation caused by deep intronic variants in cancer genomics
- Benchmarking and validating other splicing prediction tools in computational biology research
- Automating variant annotation in large-scale population genomics studies
Pros
- High Predictive Accuracy: Validated in peer-reviewed research (Cell 2019) with strong performance detecting known pathogenic splice variants, establishing it as a gold standard in the field.
- Free for Academic Use: Pre-computed annotations and source code are freely available for academic and not-for-profit research, lowering the barrier to adoption.
- Easy Pipeline Integration: Works directly with VCF files and standard bioinformatics tooling, allowing seamless incorporation into existing NGS analysis pipelines.
Cons
- Commercial License Required: Organizations using SpliceAI for commercial purposes must obtain a paid license from Illumina, which can be a barrier for industry applications.
- Archived Repository: The GitHub repository was archived in April 2026 and is now read-only, meaning no further updates, bug fixes, or new features will be released.
- Computationally Intensive: Running predictions de novo on large variant sets requires substantial compute resources; GPU acceleration is recommended for practical runtimes.
Frequently Asked Questions
What types of variants can SpliceAI annotate?
SpliceAI can annotate all possible single-nucleotide substitutions, 1-base insertions, and 1–4 base deletions within annotated genes in the human genome.
Is SpliceAI free to use?
SpliceAI is free for academic and not-for-profit use. Commercial use requires a license from Illumina, Inc.
How do I interpret SpliceAI delta scores?
Delta scores range from 0 to 1 and represent the probability that a variant alters splicing at donor or acceptor sites. Higher scores (typically ≥0.2) indicate a higher likelihood of a splice-altering effect.
Can I use SpliceAI without running the model myself?
Yes. Pre-computed annotation files for the entire genome are available for download, allowing fast lookup without needing to run the deep learning model locally.
What is the input format for SpliceAI?
SpliceAI takes standard VCF (Variant Call Format) files as input and outputs an annotated VCF with SpliceAI delta scores added to the INFO field.
