Integrative multi-omics analysis reveals breast cancer subtypes and molecular-state trajectories from static data

Breast cancer exhibits substantial molecular heterogeneity and continuous state transitions during tumor progression. However, most available clinical samples are cross-sectional, limiting direct reconstruction of the latent molecular-state relationships from normal tissue to malignant states. Here, we proposed an integrative framework to infer breast cancer subtypes and molecular-state trajectories from static multi-omics data. Using The Cancer Genome Atlas breast cancer cohort, we constructed a latent factor model integrating DNA methylation, copy number variation, mRNA expression and protein abundance, and identified four multi-omics clusters with distinct molecular characteristics and prognostic differences. By projecting normal breast samples into the pre-trained latent space using ridge regression, we constructed a branched molecular-state landscape from the normal state toward different tumor endpoints. Along the inferred trajectories, samples, which were located further from the root, tended to show poorer clinical prognosis and increased genomic instability. Our approach provides insight into progression-associated molecular variation, advances the understanding of breast cancer heterogeneity, and generates hypotheses for future experimental and clinical validation.