Accurate indel prediction using paired-end short reads

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Accurate identification of structural variants, specifically insertions and deletions (indels), remains a significant challenge in next-generation sequencing (NGS).
• Current indel calling methods rely on scoring evidence and counter-evidence, often leading to a high rate of false positives due to manually defined thresholds.
• The presence of false positive structural variants can result in spurious biological interpretations.

Purpose of the Study:

• To develop and present a machine learning-based method for accurate indel identification.
• To reduce the false positive rate in structural variant detection.
• To distinguish true indel candidates from false ones.

Main Methods:

• A discriminative classifier was developed using features from split read alignment profiles.
• The classifier was trained on a dataset of true and false indel candidates validated by Sanger sequencing.
• The method was applied to paired-end Illumina reads from 80 Arabidopsis thaliana genomes within the 1001 Genomes Project.

Main Results:

• The machine learning method effectively distinguishes true indel candidates from false positives.
• The approach significantly reduces the false positive rate in indel calling.
• Demonstrated utility in a large-scale genomic dataset (1001 Genomes Project).

Conclusions:

• Indel classification is a crucial step for minimizing false positive candidates in NGS data.
• Failure to classify indels accurately can lead to incorrect biological conclusions.
• The developed software is publicly available for use in genomic research.