DeepSSV: detecting somatic small variants in paired tumor and normal sequencing data with convolutional neural...

Area of Science:

• Genomics and Bioinformatics
• Computational Biology
• Cancer Research

Background:

• Detecting somatic mutations in paired tumor and normal sequencing data is crucial for cancer research.
• Existing somatic small variant callers often use limited information and handle biological/technological noise inconsistently, leading to divergent results.

Purpose of the Study:

• To develop a deep learning-based tool, DeepSSV, to overcome the limitations of current somatic mutation callers.
• To improve the accuracy and comprehensiveness of somatic small variant detection using convolutional neural networks (CNNs).

Main Methods:

• Developed DeepSSV, a tool employing a CNN model for somatic small variant detection.
• Created a spatially oriented representation of read alignments around candidate somatic sites for CNN processing.
• Integrated mapping information of reference and variant allele-supporting reads from tumor and normal samples in pileup format.

Main Results:

• DeepSSV effectively captures sequence dependencies and identifies context-based sequencing artifacts.
• Benchmarking on simulated and real tumor data demonstrated DeepSSV's superior performance.
• DeepSSV achieved a higher overall F1 score compared to state-of-the-art somatic callers.

Conclusions:

• DeepSSV offers a powerful deep learning approach for accurate somatic mutation detection.
• The tool's ability to process rich alignment information enhances its capability in identifying true somatic variants.
• DeepSSV represents a significant advancement over existing methods for somatic small variant calling.