DeepSNP: An End-to-End Deep Neural Network with Attention-Based Localization for Breakpoint Detection in...

Area of Science:

• Genomics
• Bioinformatics
• Artificial Intelligence

Background:

• Accurate genome-wide testing is crucial for clinical decisions in diseases like cancer.
• Classical bioinformatics tools for genomic analysis, such as Rawcopy, are often manual, time-consuming, and costly.
• These limitations hinder the timely delivery of genomic test results to physicians.

Purpose of the Study:

• To investigate the potential of deep learning algorithms to improve genome-wide analysis using single-nucleotide polymorphism array (SNPa) data.
• To develop and validate a novel deep neural network (DNN) for enhanced genomic breakpoint detection.

Main Methods:

• Development and application of DeepSNP, a novel deep neural network (DNN).
• Training and validation using a manually curated dataset of 50 SNPa analyses as a truth set.
• Comparison of DeepSNP performance against established neural network models and the Rawcopy algorithm.

Main Results:

• DeepSNP successfully learned from SNPa data to classify the presence or absence of genomic breakpoints with high precision and recall.
• The model demonstrated the ability to pinpoint genomic breakpoints, even without exact location data during training, using a localization unit.
• DeepSNP outperformed or matched existing methods in detecting genomic breakpoints.

Conclusions:

• Deep learning architectures, exemplified by DeepSNP, show significant potential for learning from genomic SNPa data.
• DeepSNP offers a more precise and efficient method for detecting genomic breakpoints compared to traditional algorithms.
• Further adaptation of DNNs for copy-number variation (CNV) detection in SNPa and other genomic data types is encouraged.