Bi-stream CNN Down Syndrome screening model based on genotyping array.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Down syndrome (DS) is a genetic disorder caused by chromosome 21 abnormalities, affecting approximately 1 in 1,000 births globally.
• Current prevention strategies for DS focus on screening and early detection due to the lack of effective cures.
• Genomic micro-duplications and dosage imbalances on chromosome 21 are primary causes of DS, leading to various abnormalities.

Purpose of the Study:

• To develop and evaluate a deep learning model for the accurate screening and prediction of Down syndrome (DS).
• To analyze genomic data using advanced machine learning techniques for improved DS detection.

Main Methods:

• Utilized deep learning, specifically a bi-stream convolutional neural network (CNN) architecture, to analyze Illumina genotyping array data.
• Converted SNP intensity data into chromosome SNP maps for image-based input into the CNN model.
• Developed a nine-layer, two-branch CNN model that merges branches for comprehensive genomic pattern analysis.

Main Results:

• The bi-stream CNN model demonstrated high average accuracies of 99.3%, with minimal false-positive and false-negative rates.
• Feature map visualization revealed genomic patterns and correlated SNP variations specific to human DS.
• Comparative analysis showed superior performance of the bi-stream CNN over single-stream CNNs and traditional machine learning models.

Conclusions:

• The bi-stream CNN effectively learns local genomic features and regional patterns across chromosomes for DS detection.
• The model's high performance and interpretable feature maps offer significant potential for disease prediction and medical practice.
• Findings provide insights for developing targeted gene therapies and advancing genomic medicine for Down syndrome.