Deep Learning-Driven Multi-Objective Clustering for Neuroblastoma Omics Data with NSGA-II

Neuroblastoma is a complex pediatric cancer with high molecular heterogeneity, making accurate subtyping and prognosis prediction challenging. Traditional clustering methods struggle with the high dimensionality of multi-omics data, leading to suboptimal stratification of diseses. This study presents a deep learning-assisted multi-objective clustering framework integrating autoencoders for dimensionality reduction with the Non-Dominated Sorting Genetic Algorithm II (NSGA-II) to enhance clustering accuracy and biological relevance. The autoencoder extracts essential features while minimizing noise, generating a compact latent representation of multi-omics data. NSGA-II then optimizes intra-cluster compactness and inter-cluster separation to improve patient stratification. The framework was validated on publicly available neuroblastoma datasets encompassing genomics, transcriptomics, and epigenomics data. Comparative analysis showed that our approach improves clustering accuracy by 15% and enhances survival prediction by 20% over conventional clustering methods. Kaplan–Meier survival analysis further confirmed the clinical relevance of the identified clusters. The proposed framework provides a scalable and efficient method for high-dimensional, heterogeneous multi-omics data analysis, contributing to improved precision medicine strategies for neuroblastoma patients.