A Machine Learning approach combining omics data for Alzheimer’s Disease analysis

Alzheimer’s disease (AD) is a complex neurological disorder whose underlying mechanisms remain elusive to this day. Molecular biology methodologies, especially techniques like single-cell RNA sequencing (scRNA-seq), offer unparalleled granularity in deciphering the disease’s cellular intricacies. However, despite the potential of scRNA-seq, comprehensive machine-learning analyses are yet to be fully harnessed. Emphasizing the multi-omics machine-learning-based approaches, which integrate diverse single-cell omics datasets, could highlight novel therapeutic targets and deepen our understanding of AD’s intricate nature. In this work, we propose a machine-learning-based method to embed gene expression into a protein interaction graph. Specifically, we model each interaction with a regression model on the participating genes and we use the R2 score as edge weight. Our aim is to detect parts of the PPI graph that differentiate between control and disease conditions. Application on a scRNA-seq AD dataset managed to identify interactions forming small subgraphs, which consisted of genes involved with biological processes related to neurons.