LoQuM

LoQuM recalibrates mapping quality scores for Illumina short-read alignments using a logistic regression model trained on simulated reads to improve the reliability of mapping quality assessments.

Key Features:

• Machine learning calibration: Uses a logistic regression model trained on simulated reads to predict true mapping quality for short-read alignments.
• Feature set: Utilizes features derived from sequencer-reported base quality scores and alignment-specific statistics, including number of matches, mismatches, deletions, mapping quality score from the original aligner, and number of mappings.
• Rescue of zero-quality mappings: Reassigns nonzero calibrated mapping quality to alignments originally assigned zero by alignment tools.
• Improved variant signal: Recalibrated mapping qualities increase the precision of identifying genomic variants such as single nucleotide polymorphisms (SNPs).

Scientific Applications:

• Variant calling accuracy: Provides more reliable mapping quality inputs to variant callers, improving SNP and small variant detection.
• Structural variation analysis: Enhances assessment of mapping reliability in contexts relevant to structural variant discovery.
• Population genetics and evolutionary studies: Improves confidence in mapping-based measures used for genetic diversity and evolutionary inference.
• Disease association studies: Increases fidelity of mapped reads used in analyses linking genomic variants to disease phenotypes.

Methodology:

Logistic regression trained on simulated reads using features from sequencer-reported base quality scores and alignment statistics (matches, mismatches, deletions, original aligner mapping quality, and number of mappings).