This study presents an integrated approach for discovering and clustering significant genes associated with neurodegenerative diseases using a combination of variance-based filtering, Long Short-Term Memory (LSTM) neural networks, and enrichment analysis. By first filtering genes based on their variance across samples, we retain genes with high biological relevance. These selected genes are then used to train an LSTM model, capturing complex interactions between gene expression patterns. Using Uniform Manifold Approximation and Projection (UMAP) for dimensionality reduction and K-Means for clustering, we group genes with similar expression profiles. The optimal number of clusters is determined using the Silhouette Score. To refine the model, we introduce a feedback loop where cluster labels are fed back into the LSTM as additional features, enhancing the model’s ability to detect significant gene associations. We compare various network architectures to assess their impact on clustering performance. Finally, enrichment analysis reveals key biological pathways, such as immune regulation and protein signaling, related to Parkinson’s disease, Alzheimer’s disease, and amyotrophic lateral sclerosis (ALS). Our approach demonstrates the potential of machine learning and clustering to uncover meaningful gene associations, offering insights into the molecular mechanisms underlying neurodegenerative diseases.