A probabilistic approach for SNP discovery in high-throughput human resequencing data

Area of Science:

• Genomics
• Bioinformatics
• Machine Learning

Background:

• High-throughput sequencing generates vast amounts of data, necessitating accurate variant identification.
• Identifying heterozygous sites, especially in low-coverage regions, remains a significant challenge in genetic studies.

Purpose of the Study:

• To develop and evaluate a novel machine learning strategy for accurate identification of heterozygous sites in single individuals.
• To provide confidence scores for combining sequencing data from multiple samples and identifying poorly sequenced regions.

Main Methods:

• A machine learning approach generating a heterozygosity score for each chromosomal position.
• Evaluation on HapMap gold standard data and de novo sequencing data.
• Comparison with high-coverage 1000 Genomes sequencing data.

Main Results:

• Achieved 98.5% accuracy in classifying heterozygous/homozygous sites on HapMap data.
• ProbHD identified 93% of heterozygous sites with <5% false call rate in de novo data.
• Demonstrated >99.9% overall genotype call agreement and ~90% heterozygote call agreement with 1000 Genomes data.

Conclusions:

• Machine learning-based sequence analysis can alleviate systematic biases in high-throughput resequencing.
• The ProbHD method offers a robust solution for accurate DNA variant extraction, particularly for heterozygous sites.
• This approach enhances the reliability of large-scale human genomic resequencing studies.