PredPSI-SVR

PredPSI-SVR predicts changes in percent spliced in (ΔPSI) caused by genetic variants, focusing on exon skipping events to assess effects on alternative splicing.

Key Features:

• Support Vector Regression (SVR): Employs SVR to model and predict ΔPSI from sequence-derived features.
• Sequence-derived features: Extracts 42 distinct features from the exon sequence and its flanking regions.
• Greedy feature selection: Applies a greedy algorithm to identify the eight most contributory features for prediction.
• Event focus: Targets exon skipping events and quantifies impact using percent spliced in (ΔPSI).
• CAGI/vex-seq validation: Validated in the "vex-seq" challenge of the 5th Critical Assessment of Genome Interpretation (CAGI).
• Performance metrics: Achieved PCC = 0.570 in 10-fold cross-validation on training data and PCC = 0.566 in the blind test, ranking second in the challenge.
• Synonymous mutation prioritization: Capable of prioritizing deleterious synonymous mutations that affect splicing.

Scientific Applications:

• Variant effect prediction: Predicts ΔPSI changes caused by genetic variants affecting exon skipping.
• Variant prioritization: Prioritizes synonymous and other variants with potential deleterious effects on splicing.
• Disease research: Assesses impacts of splicing-altering variants relevant to disease etiology, including cancer.
• Benchmarking: Serves as a model for method evaluation in challenges such as CAGI vex-seq.

Methodology:

Uses support vector regression trained on 42 sequence-derived features from exons and flanking regions with greedy selection of eight features, evaluated by 10-fold cross-validation and blind testing in the CAGI "vex-seq" challenge.