SNooPer

SNooPer applies Random Forest classification to identify somatic variants in cancer genomes from low-pass next-generation sequencing (NGS) data.

Key Features:

• Random Forest classification: Employs Random Forest models to distinguish genuine somatic variants from sequencing errors.
• Data-specific model training: Trains a dataset-specific model using a subset of variant positions with known true class (genuine variation versus sequencing error).
• Adaptation to dataset characteristics: Models adapt to the specific technical and biological characteristics of each sequencing dataset.
• Low-pass sequencing support: Designed to operate on low-pass whole-exome and whole-genome sequencing data.
• Low VAF detection: Optimized to improve identification of variants present at low variant allele frequencies (VAFs).
• Robustness to technical artifacts: Accounts for sequencing depth variability, alignment errors, and systematic technical biases.
• Improved specificity and sensitivity: Demonstrates higher specificity and sensitivity compared to benchmarked somatic callers in reported evaluations.
• Cost-efficiency implication: Enables accurate somatic variant calling from low-pass data, which can reduce overall sequencing requirements.

Scientific Applications:

• Somatic variant identification in cancer genomes: Detection of somatic single-nucleotide variants and small events from tumor sequencing data.
• Low-pass WES/WGS studies: Variant calling in studies using low-pass whole-exome or whole-genome sequencing.
• Analysis of heterogeneous tumors: Identification of variants in samples with tumor heterogeneity and low tumor purity.
• Pediatric oncology cohorts: Applied to childhood acute lymphoblastic leukemia cohorts for benchmarking and discovery.

Methodology:

Random Forest classification models are trained on a data-specific training set consisting of variant positions with known true class (genuine variant versus sequencing error) and then applied to classify somatic variants in the dataset.