SQUIRLS

SQUIRLS prioritizes splice-altering genetic variants by computing information content (IC) changes for wildtype and variant sequences at canonical and cryptic donor and acceptor splice sites and applying machine-learning classifiers to predict their impact.

Key Features:

• Information Content Calculation: Calculates information content (IC) for wildtype and variant sequences at canonical and cryptic splice sites to assess changes in splicing regulatory signals.
• Feature Generation: Produces a concise, interpretable feature set including exon length, disruptions of the AG exclusion zone, and sequence conservation metrics.
• Random-Forest Classifiers: Employs two random-forest classifiers, one trained for donor splice sites and one for acceptor splice sites, to capture site-specific regulatory mechanisms.
• Logistic Regression Integration: Combines the outputs of the donor and acceptor classifiers using logistic regression to produce a unified score for variant prioritization.
• Performance and Speed: Demonstrates improved ranking accuracy in simulated exomes relative to prior methods and offers faster runtimes suitable for high-throughput analyses.
• Tabular Output: Produces tabular output files intended for integration into exome and genome diagnostic pipelines.

Scientific Applications:

• Genetic Diagnostics: Prioritizes splice-altering variants, including nucleotide changes outside the conserved AG/GT dinucleotides, for clinical variant interpretation.
• Assessment of Noncanonical and Cryptic Sites: Extends variant assessment to cryptic splice sites and intronic changes beyond canonical dinucleotides to identify disease-associated splicing alterations.
• Functional Genomics and Personalized Medicine Research: Provides interpretable features to support research into the functional implications of splice variants in genomics and personalized medicine studies.

Methodology:

Computes information content for wildtype and variant sequences at canonical and cryptic splice sites, generates features (exon length, AG exclusion zone disruptions, sequence conservation), applies two random-forest classifiers (donor and acceptor), combines classifier outputs with logistic regression, and evaluates performance by rank analysis in simulated exomes.