CMVAE: Contrastive Supervision Improves Single-Omic Latent Manifold for Breast Cancer Survival Prognosis

CMVAE improves breast invasive carcinoma survival prognosis from single-omic data, reaching AUROC 0.7397 {+/-} 0.0239 on TCGA-BRCA using leak-proof five-fold stratified cross-validation.

• Why it matters: Single-omic representation learning is hindered by missingness and limited overlap across modalities, and standard reconstruction-focused latent spaces may not reflect downstream survival outcomes.
• What they did: The authors train a Contrastive Masked Denoising Variational Autoencoder with stochastic feature masking to infer corrupted molecular inputs and contrastively align patient survival trajectories.
• The result: In TCGA-BRCA, CMVAE yields highly significant risk stratification between low- and high-risk groups (p < 0.0001) and outperforms six multi-omics survival baselines by at least 3%.

AI summary of the abstract · may contain errors · not medical advice · read the paper.