Detection of chromosome structural variation by targeted next-generation sequencing and a deep learning application

Area of Science:

• Oncology
• Genomics
• Bioinformatics

Background:

• Molecular testing, particularly targeted next-generation sequencing (NGS), is crucial for cancer diagnosis.
• Detecting structural variations (SV) like 1p/19q co-deletion via targeted NGS presents challenges.
• Accurate identification of molecular alterations is vital for glial tumor classification.

Purpose of the Study:

• To develop and validate a deep learning (DL) model for detecting 1p/19q co-deletion using targeted NGS.
• To assess the utility of DL in SV detection for brain tumor molecular profiling.
• To improve the accuracy of glial tumor classification through enhanced molecular analysis.

Main Methods:

• Development of an ensemble 1-dimensional convolution neural network (DL model).
• Application of the DL model to targeted NGS data from 61 brain tumors, including 19 oligodendroglial tumors.
• External validation of the DL model using The Cancer Genome Atlas (TCGA) low-grade glial tumor data (n=427).

Main Results:

• Manual review confirmed 1p/19q co-deletion in all 19 oligodendroglial tumors.
• The DL model achieved perfect detection (AUC=1) in the testing set.
• The DL model demonstrated reproducible results (AUC=0.9652) in the external validation set, despite different data generation platforms.

Conclusions:

• Targeted NGS with a cancer gene panel is a viable method for glial tumor classification.
• Deep learning models can be effectively integrated for robust SV detection in NGS data.
• This approach enhances the accuracy and reliability of molecular diagnostics in neuro-oncology.