CNV-P

CNV-P refines copy number variation (CNV) predictions from existing CNV detection tools by applying a machine-learning classifier trained on read depth (RD), split reads (SR), and read pair (RP) signals to reduce false positives and improve precision and recall in genomic sequencing data.

Key Features:

• Post-processing filtering: Filters erroneous CNV predictions from the outputs of conventional CNV detection tools to reduce false-positive rates.
• Machine-learning classifier: Uses a trained classifier that leverages genomic signal features rather than raw callers' scores.
• Signal features: Defines and uses read depth (RD), split reads (SR), and read pair (RP) signals as features for model training.
• Empirical performance: Demonstrates over 90% precision and 85% recall in classification on multiple real biological sequencing datasets.
• Robustness: Maintains performance across various CNV sizes and sequencing platforms.

Scientific Applications:

• Genetic disorder research: Improves the reliability of CNV calls used to investigate CNV contributions to genetic diseases.
• Clinical diagnostics: Enhances confidence in CNV detection for clinical genomic interpretation by reducing false positives.
• Genomic studies: Increases accuracy of CNV catalogs in basic research and population sequencing projects.

Methodology:

CNV-P post-processes CNV calls from conventional CNV detection tools using a machine-learning classifier trained on read depth (RD), split reads (SR), and read pair (RP) signal features, with training and evaluation performed on multiple real biological sequencing datasets.