INTEGRATED COMPUTATIONAL ANALYSIS FOR DEEP LEARNING BASED NEURODEGENERATIVE DISORDER DIAGNOSIS AND BIOMARKER DISCOVERY

Abstract

Aim: This paper will develop an integrative computational framework to study neurodegenerative diseases in transcriptomic analysis of ALS and DL-based classification of AD by utilizing structural MRI. This aims at identifying the major molecular biomarkers and establishing a precise, interpretable diagnostic model.

Background: NDDs like ALS and AD are demonstrated by progressive neuronal loss and complex underlying mechanisms involving neuroinflammation, protein aggregation, and cellular dysfunction. Whereas transcriptomic research can reveal insights into non-invasive and early detectable disease dysregulation at the gene level, neuroimaging in combination with artificial intelligence can provide insights into how diseases can be detected at an early and non-invasive stage of disease progression. Nevertheless, the inability to integrate molecular and imaging methods restricts a full picture of disease pathology.

Methodology:Transcriptomic analysis was done on the GSE10953 dataset that was provided by the Gene Expression Omnibus. First, the differential gene expression was analyzed using the Limma package, then key hub genes were determined by functional enrichment and protein protein interaction (PPI) network analysis. At the same time, a CNN was developed to classify the different Alzheimer's disease stages based on the publicly available datasets which were presented in the form of MR images. A model was used for preprocessing methods and data augmentation and evaluated using the standard performance metrics. Grad-CAM was used to make the models more interpretable, visualizing discriminative brain areas.

Findings: The transcriptomic analysis revealed that 1216 differentially expressed genes, including SKP1A, CALM13, RPS14, H3C7, and CD44, were associated with neuroinflammation, ubiquitin proteasome dysfunction, epigenetic dysregulation, and ribosomal stress. The overall accuracy of the deep learning model was 94 per cent, characterized by high values in precision, recall, F1 score and ROC-AUC. Grad-CAM visualizations showed that the model targets clinically relevant brain areas that are related to neurodegeneration. An integrated approach to molecular and imaging discoveries reveals complementary insight into disease pathogenesis and the potential of complementary bioinformatics and artificial intelligence to enhance diagnosis and biomarker discovery in neurodegenerative diseases.