This study presents a comparative analysis of three dimensionality reduction techniques—MOFA (Multi-Omics Factor Analysis), IntNMF (Integrative Non-negative Matrix Factorization), and VAE (Variational Autoencoder)—applied to breast cancer multi-omics data from the TCGA-BRCA project. The integration of omics layers—including gene expression, protein expression, copy number variations, and mutations—was combined with key clinical variables to evaluate the performance of latent representations in both classification and clustering tasks. Major challenges such as high dimensionality and severe class imbalance were addressed through oversampling and undersampling strategies. Each method was evaluated for its effectiveness in predicting clinical outcomes and identifying meaningful molecular patterns. MOFA offered biologically interpretable and stable representations, IntNMF produced compact structures, and VAE yielded well-separated latent spaces. Enrichment analysis confirmed the relevance of extracted features, reinforcing the utility of latent factor models for robust multi-omics integration in breast cancer research.