tarSVM

tarSVM improves the accuracy of variant calling from microfluidic PCR-based targeted sequencing by applying a Support Vector Machine classifier and a normalized allele dosage test to filter false positives arising from PCR amplification and high duplication rates.

Key Features:

• Support Vector Machine (SVM) Integration: tarSVM employs a Support Vector Machine classifier to distinguish true variants from false positives using variant-derived features.
• Normalized Allele Dosage Test: It applies a normalized allele dosage test as a scoring metric to assess variant quality and allele frequency accuracy.
• Reference Data Integration: The pipeline incorporates variant information from the 1000 Genomes Project and the Exome Aggregation Consortium (ExAC) to augment training data with high-confidence variants.
• GATK Feature Utilization: tarSVM synthesizes variant features derived from the Genome Analysis Toolkit (GATK) to leverage established variant calling metrics.
• PCR Duplication Handling: The method specifically addresses high duplication rates inherent to microfluidic PCR-based targeted sequencing to reduce PCR-induced false positive calls.
• Reported Performance: In comparative studies across cohorts (n = 474 and n = 1152) and a validation cohort (n = 75), tarSVM reported accuracies of 84.5% and 73.3% in the two cohorts and a 5% false discovery rate in the validation cohort.

Scientific Applications:

• Targeted gene-set sequencing: Improves variant call accuracy for microfluidic PCR-based targeted sequencing of discrete gene sets.
• Monogenic disease screening: Enhances reliability of variant detection for cost-effective screening of monogenic diseases and reduces reliance on Sanger validation.
• Downstream variant analysis: Produces higher-confidence variant calls for downstream analyses that depend on accurate allele frequency and quality metrics.

Methodology:

Preprocessing by read trimming; alignment of trimmed reads to a reference genome using established alignment tools; variant calling from aligned reads; and variant quality filtering using the Support Vector Machine classifier and the normalized allele dosage test.