Pangolin: AI-Powered Splice Site Prediction Now Available on Vecura

What is Pangolin?

Pangolin is a deep-learning model that predicts RNA splicing patterns directly from DNA sequence. It uses a dilated residual convolutional neural network architecture trained on tissue-specific RNA-seq data from GTEx to capture the long-range sequence context (up to ±5000 bases) that determines splice-site recognition. The model ships as an ensemble of four tissue-specific models (Heart, Liver, Brain, and Testis), each providing both probability-of-splicing and splice-site usage predictions.

It helps users quantify the impact of genetic variants on splice-site strength and identify pathogenic splicing mutations. It is especially useful for distinguishing which of the millions of genetic variants in a human genome alter RNA splicing, with particularly strong performance on intronic variants located far from exon boundaries.

What can users do with Pangolin on Vecura?

With Pangolin on Vecura, users can:

• Score genetic variants from VCF or CSV files to predict their impact on splice-site strength, receiving delta-scores that quantify predicted gains or losses
• Generate per-base splice probability tracks across custom DNA sequences for multiple tissue types (Heart, Liver, Brain, Testis)
• Identify the largest predicted splice gains and losses near each variant, with position and magnitude information
• Filter results using customizable parameters like distance windows, score cutoffs, and annotation masking to focus on biologically relevant predictions

What the output means

The output provides delta-scores for each variant indicating the predicted change in splice-site strength (positive values for gains, negative values for losses), along with structured data showing the position and magnitude of the largest splice effects. For custom sequences, per-base probability tracks show splice-site usage patterns across tissue types.

This output should be used to support scientific decision making. It does not replace experimental validation.

Why this matters

Accurate interpretation of genetic variants remains one of the most significant challenges in precision medicine. While exome and genome sequencing have become routine, the functional impact of most variants—particularly those in non-coding regions or deep within introns—remains unclear. Aberrant RNA splicing is a major disease mechanism, accounting for a substantial proportion of pathogenic variants in Mendelian disorders and cancer. Traditional splicing prediction tools often struggle with variants located far from canonical splice sites, missing cryptic splice sites and tissue-specific effects.

Pangolin addresses these limitations by leveraging tissue-specific RNA-seq data and a deep learning architecture designed to capture long-range sequence dependencies. Benchmarks demonstrate superior accuracy compared to earlier tools like SpliceAI, particularly for intronic variants distant from exon boundaries. By making Pangolin accessible through Vecura's guided workflow, researchers and clinicians can integrate advanced splicing predictions into their variant interpretation pipelines without managing complex GPU infrastructure or Python dependencies, accelerating the identification of disease-causing splicing mutations.

• Developed by: Zeng & Li
• Source: Official GitHub repository (https://github.com/tkzeng/Pangolin)
• Reference: Zeng & Li, Genome Biology 2022 (https://doi.org/10.1186/s13059-022-02664-4)