Scalable Big Data And Neural Network Architectures For Integrative Gene Expression Analysis In Parkinson’s And Alzheimer’s Disease Research

Abstract

Parkinson and Alzheimer are the examples of neurodegenerative diseases, characterised by complicated molecular processes that include neuroinflammation, mitochondrial abnormalities, synaptic imbalances, and oxidative stress. The increasing accessibility of transcriptomic scale data sets requires scalable computational infrastructure with the potential of an integrative analysis of heterogeneous cohorts. The research will suggest a scalable paradigm of big data that will integrate the neural network-based architectures in the study of integrative gene expression in PD and AD research. Published transcriptomic data at Gene Expression Omnibus (GEO) were aggregated and batch-corrected and harmonised to create a cross-disease expression matrix of 1,842 samples. Deep neural models were used alongside a differential expression analysis in order to determine shared and disease-specific biomarkers. Several neural architecture types were analysed such as multilayer perceptrons (MLP), convolutional neural networks (CNN) and stacked autoencoders. The best architecture had AUC of 0.94 in AD classification, and 0.91 in PD classification, which was again supported by cross-cohort validation. Convergent dysregulation in mitochondrial oxidative phosphorylation, pathways of microglial activation and synaptic transmission were identified through integrative pathway enrichment. This scalable framework indicates how it is possible to use deep learning-based integrative transcriptomic analysis to identify molecular overlap and divergence between key neurodegenerative diseases.