ALS

ALS predicts amyotrophic lateral sclerosis (ALS) status from genotype data using deep convolutional neural networks that model promoter and regulatory regions and capture non-additive genetic interactions.

Key Features:

• Deep Learning Framework: Employs convolutional neural networks (CNNs) adapted to the biological structure of genomic data, applying convolutions selectively to relevant genomic regions.
• Two-Step Approach: First identifies promoter regions likely associated with ALS by analyzing regulatory elements that harbor disease-associated variants, then classifies individuals using a deep CNN trained on genotypes within those regions.
• Non-Additive Genetic Interactions: Captures complex, non-linear and non-additive interactions between genetic variants that may contribute to ALS heritability.
• Performance and Validation: Demonstrated superior performance compared to other classification methods when validated on the Dutch cohort of the Project MinE dataset.
• Scalability for Whole-Genome Data: Architecture is optimized for processing large-scale whole-genome genotype datasets for population-level analyses.

Scientific Applications:

• Disease Prediction: Classifies individuals as ALS cases or controls from genotype data to support risk assessment and early identification efforts.
• Genetic Research: Facilitates identification of promoter and regulatory regions associated with ALS, informing studies of disease-associated variants.
• Methodological Advancement: Provides a model for integrating deep learning, selective convolutions, and region-based genotype analysis in studies of complex genetic diseases.

Methodology:

Data preparation using synthetic datasets and real genotype data from large cohorts; design of CNN architectures that respect genomic structure and apply convolutions selectively to promoter/regulatory regions; and model training and testing on identified promoter regions with evaluation against traditional classification methods.