Amyotrophic lateral sclerosis diagnosis using machine learning and multi-omic data integration.

Area of Science:

• Neuroscience
• Genetics
• Computational Biology

Background:

• Amyotrophic Lateral Sclerosis (ALS) presents significant genetic and molecular heterogeneity, with many underlying genetic factors remaining unknown.
• The complexity of ALS necessitates personalized medicine approaches for improved diagnosis, prognosis, and treatment.
• Current understanding of ALS pathogenesis is limited by the heterogeneity and incomplete knowledge of genetic factors.

Purpose of the Study:

• To develop a comprehensive, machine learning-facilitated, multi-omic model for a deeper understanding of Amyotrophic Lateral Sclerosis (ALS).
• To integrate gene expression profiles and rare genomic variants to identify key genes and pathways involved in ALS.
• To create a model that exposes intricate genotype-phenotype interconnections for improved ALS diagnosis and interpretation.

Main Methods:

• Unsupervised clustering was applied to gene expression profiles to identify 9,847 ALS-associated genes.
• These genes were integrated with 7,699 genes harboring rare genomic variants, creating a dataset of 17,546 genes.
• A Variational Autoencoder was employed to develop the Multi-Omics for ALS (MOALS) model, distilling complex biomedical information.

Main Results:

• The MOALS model successfully elucidated pivotal ALS signaling pathways.
• MOALS demonstrated superior performance compared to single-omic machine learning models (SNV and RNA expression).
• Accuracy was enhanced by 1.7% over SNV-based models and 6.2% over RNA expression-based models.

Conclusions:

• Analyzing biological system relationships provides heuristic insights into ALS mechanisms.
• The MOALS model offers a more accurate and interpretable approach to ALS diagnosis.
• This multi-omic strategy advances the understanding of genotype-phenotype correlations in ALS.