CNNdel: Calling Structural Variations on Low Coverage Data Based on Convolutional Neural Networks.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Next-generation sequencing (NGS) has led to numerous structural variation (SV) detection methods.
• Existing SV callers struggle with accuracy on low-coverage sequences, resulting in many false positives.
• Combining results from multiple tools improves sensitivity but not accuracy for low-coverage data.

Purpose of the Study:

• To develop an accurate method for identifying deletion structural variations from paired-end reads.
• To address the challenge of high false positive rates in low-coverage sequencing data.
• To leverage machine learning for improved SV detection.

Main Methods:

• CNNdel gathers SV candidates from multiple existing tools.
• Features are extracted from aligned BAM files at candidate positions.
• Convolutional neural networks (CNNs) are trained on labeled data to classify true vs. false SV candidates.
• The approach was tested on low-coverage genomes from the 1000 Genomes Project.

Main Results:

• CNNdel effectively identifies deletion structural variations.
• The method demonstrates high accuracy even with low-coverage NGS reads.
• CNNdel significantly reduces false positives compared to existing approaches.
• Convolutional neural networks automatically prioritize important SV features.

Conclusions:

• CNNdel offers an effective solution for accurate deletion detection in low-coverage genomic data.
• The study highlights the efficacy of CNNs in improving SV calling accuracy by reducing false positives.
• This approach enhances the reliability of structural variation analysis in genomics research.