LSnet: detecting and genotyping deletions using deep learning network.

Area of Science:

• Genomics and Bioinformatics
• Computational Biology
• Machine Learning in Biology

Background:

• Structural variations (SVs), particularly deletions, play a significant role in biological functions and disease.
• Accurate detection and genotyping of deletions are crucial for understanding genomic complexity.
• Existing methods face challenges due to genome complexity and alignment intricacies.

Purpose of the Study:

• To develop and present LSnet, a novel deep learning-based approach for detecting and genotyping deletions.
• To improve the accuracy and efficiency of deletion detection using advanced sequencing data.

Main Methods:

• LSnet utilizes a deep learning network, combining convolutional neural networks (CNNs) and gated recurrent units (GRUs) with an attention mechanism.
• The method processes accurate long reads (HiFi or combined error-prone long and short reads) by dividing the reference genome into sub-regions.
• Nine features are extracted per sub-region, followed by sequential learning to identify deletion signatures and a heuristic algorithm for precise localization.

Main Results:

• LSnet demonstrates superior performance in deletion detection and genotyping compared to existing methods.
• The approach achieves a high F1 score, indicating significant improvements in accuracy and reliability.
• The deep learning architecture effectively learns complex genomic features indicative of deletions.

Conclusions:

• LSnet offers a powerful and accurate solution for detecting and genotyping deletions, addressing a key challenge in structural variation analysis.
• The integration of deep learning with advanced sequencing technologies holds great promise for advancing genomic research.
• The source code is publicly available, facilitating further research and application of the LSnet method.