CN-RNN: a Deep Learning Framework for Copy Number Variation Detection with Exome Sequencing Data

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Copy number variations (CNVs) are significant structural genomic alterations linked to numerous human diseases.
• Accurate CNV detection from whole-exome sequencing (WES) data is essential for clinical genetics and population studies.
• Current WES-based CNV detection methods exhibit limitations, including high false-positive rates and poor recall for short variants, with deep learning models not fully leveraging complementary genomic information.

Purpose of the Study:

• To develop and present CN-RNN, a novel deep learning-based tool for accurate CNV detection from WES data.
• To improve the accuracy and recall of CNV detection, particularly for short variants, by integrating local and region-level genomic features.
• To provide a scalable and accurate CNV profiling tool for WES-based studies, facilitating broader applications in population and clinical research.

Main Methods:

• Developed CN-RNN, a deep learning model integrating a bidirectional long short-term memory (BiLSTM) branch for local depth changes and a multi-layer perceptron (MLP) branch for region-level metadata (GC content, mappability, exon length).
• Trained CN-RNN using the Autism Sequencing Consortium (ASC) parent-child trio cohort, enforcing Mendelian inheritance rules for high-quality training data.
• Evaluated CN-RNN performance across three independent datasets to compare its accuracy against existing WES-based CNV callers and deep learning methods.

Main Results:

• CN-RNN demonstrated superior performance compared to existing WES-based CNV callers and other deep learning methods across multiple independent datasets.
• The model effectively captures local depth variations and contextual dependencies using the BiLSTM branch, while the MLP branch incorporates crucial region-level genomic features.
• Achieved improved accuracy and recall in CNV detection, addressing limitations of previous WES-based approaches.

Conclusions:

• CN-RNN offers a significant advancement in CNV detection from WES data, providing a scalable and accurate tool.
• The integration of complementary genomic information in CN-RNN enhances its performance and reliability for clinical and population genetic studies.
• CN-RNN is poised to broaden the application and impact of CNV analysis in genetic research.