mitoDataclean

mitoDataclean detects cross-contamination in mitochondrial DNA (mtDNA) next-generation sequencing (NGS) data using machine learning to distinguish genuine from contamination-derived variants for applications in aging and cancer research.

Key Features:

• Machine Learning-Based Approach: Employs a random-forest algorithm to detect and quantify cross-contamination in mtDNA NGS data, enabling discrimination beyond haplogroup-level phylogeny.
• Haplotype-Level Sensitivity: Optimized to detect subtle haplotype-level differences to distinguish genuine mtDNA variants from contamination-derived variants.
• Comprehensive Optimization: Trained and optimized using simulated mixtures with small haplogroup distances and low polymorphic differences.
• High Sensitivity and Accuracy: Demonstrated AUC values of 0.91 for Western datasets and 0.97 for private sequencing contamination data in simulated evaluations.
• Versatility Across Populations and Samples: Showed robust performance across diverse datasets, indicating applicability to different populations and contamination sources.

Scientific Applications:

• Oncology: Detects and quantifies mtDNA cross-contamination in cancer sequencing studies to improve reliability of mitochondrial variant interpretation and analyses of cancer pathogenesis.
• Gerontology: Ensures mtDNA data integrity in aging research and studies of mitochondrial function.

Methodology:

Trains a random-forest model on simulated mixtures reflecting realistic contamination scenarios, including small haplogroup distances and low polymorphic differences, and evaluates performance on private and public NGS datasets.