Splice Variant Prediction - Usage Guide

Overview

Predict whether a DNA variant alters mRNA splicing using sequence-based deep-learning tools - SpliceAI (10kb context CNN), Pangolin (multi-tissue), MMSplice (modular per-region), SpliceTransformer/TrASPr (tissue-aware transformers), SpliceVault (empirical 300K-RNA lookup of likely mis-splicing outcomes), and CADD-Splice. Applies the ClinGen SVI 2023 framework for ACMG/AMP variant interpretation (PVS1, PP3, BP4 evidence codes), HGVS splicing nomenclature, extended-window scoring for deep-intronic pseudoexons, and tissue-specific predictions.

Prerequisites

# Python
pip install spliceai tensorflow pyfaidx pyensembl pysam

# Pangolin (splicing) from GitHub - NOT the PyPI 'pangolin' (SARS-CoV-2 lineage tool)
pip install git+https://github.com/tkzeng/Pangolin.git
# Plus a gffutils annotation database
python3 -c "import gffutils; gffutils.create_db('gencode.v45.gff3', 'gencode.db')"

# MMSplice (specific dependencies)
pip install mmsplice

# SpliceVault: web at kidsneuro.shinyapps.io/splicevault or R/Python at github.com/kidsneuro-lab/SpliceVault

# Reference files
# - Genome FASTA (GRCh38 or GRCh37)
# - GENCODE GTF / GFF3

Quick Start

Tell your AI agent what you want to do: - "Predict SpliceAI delta scores for variants in my VCF" - "Score a panel of clinical variants for splice impact with concordance across SpliceAI, Pangolin, and MMSplice" - "Identify deep-intronic pseudoexon-creating variants using extended-window SpliceAI" - "Classify a candidate variant under ClinGen SVI 2023 splicing rules (PVS1/PP3/BP4)" - "Predict tissue-specific splicing impact with Pangolin for a brain-disease variant"

Example Prompts

Single Variant

"For variant chr17:g.7676154A>G (TP53), compute SpliceAI delta score with -D 50 and Pangolin tissue-specific predictions for brain."

VCF Annotation

"Run SpliceAI on my VCF of clinical variants with grch38 reference, then filter to variants with delta_max >= 0.2 and apply ClinGen PP3 thresholds."

Deep-Intronic

"I have an unsolved Mendelian case; re-run SpliceAI with -D 2000 to check for pseudoexon-creating deep-intronic variants."

Concordance

"Run SpliceAI + Pangolin + MMSplice on a candidate VUS panel; flag discordant predictions for RNA validation."

ASO Design

"Predict the splicing impact of occluding a specific ESE region with a 22-mer ASO using SpliceAI on the masked sequence."

Empirical Outcome

"For a canonical 5'ss-disrupting variant, query SpliceVault to predict whether it causes exon skipping vs cryptic site activation."

What the Agent Will Do

Validate variant nomenclature (HGVS) with VariantValidator or Mutalyzer
Run SpliceAI with appropriate distance window (50 default; 500-2000 for deep-intronic)
Run Pangolin if tissue-specific predictions are needed
Run MMSplice for calibrated ΔPSI in cassette-exon contexts
Query SpliceVault for empirical mis-splicing outcomes
Apply ClinGen SVI 2023 thresholds for PP3/BP4/PVS1 evidence
Recommend RNA validation for clinical-grade reporting

Tips

ClinGen SVI 2023: SpliceAI >=0.2 = PP3, <=0.1 = BP4, both applied at SUPPORTING weight only (0.5/0.8 raise SpliceAI precision but are not ClinGen evidence-strength upgrades; moderate/strong needs RNA/functional PS3 evidence)
SpliceAI default window is 50nt; extend to 500-2000 for unsolved Mendelian cases (5-15% are deep-intronic pseudoexons)
Tissue-agnostic prediction (SpliceAI) is fine for screening; use Pangolin/SpliceTransformer when disease tissue is known
RNA validation (PS3) supersedes computational evidence (PP3); run RT-PCR or RNA-seq when possible
Branchpoint variants are weakly predicted by all current tools; use BPHunter (Zhang 2022) as supplement
Concordance across SpliceAI + Pangolin + MMSplice strengthens evidence; discordance flags need experimental work
Log SpliceAI version and distance window with every clinical call
SpliceAI alone is NOT sufficient for PVS1; canonical site disruption requires gene-level LoF context

splicing-qc - MaxEntScan + library QC for confirming predicted impact
splicing-quantification - Empirical PSI from RNA-seq to validate predictions
outlier-splicing-detection - FRASER2/DROP for RNA-seq confirmation in clinical samples
variant-calling/clinical-interpretation - Broader ACMG/AMP variant interpretation
variant-calling/variant-annotation - VEP plugin integration for SpliceAI