CMVAE: Contrastive Supervision Improves Single-Omic Latent Manifold for Breast Cancer Survival Prognosis
- 1Bellini College of AI, Cybersecurity and Computing, University of South Florida
Abstract
High-dimensional omics data are vital for fore- casting patient prognosis in oncology. While deep multi-omics learning methods are increasingly popular, they suffer from data missingness and low patient overlap across modalities. Consequently, advancing single-omic representation learning re- mains a critical yet challenging research problem. Standard generative dimensionality reduction methods rely on recon- struction objectives that optimize the global manifold topology; however, overlook downstream clinical outcomes. To address these limitations, we introduce a Contrastive Masked Denois- ing Variational Autoencoder (CMVAE), a self-supervised dual- optimization framework designed to map single-omic information into a clinical outcome-aware latent space. A stochastic feature masking strategy corrupts the input omic profiles, and CMVAE is optimized to simultaneously infer missing molecular data and contrastive alignment of patient survival trajectories. We evaluate CMVAE using leak-proof five-fold stratified cross-validation on the TCGA Breast Invasive Carcinoma (BRCA) cohort. The method outperforms traditional unsupervised methods, achieving an AUROC of 0.7397 {+/-} 0.0239 on patient prognosis. Further analysis demonstrates a highly significant (p < 0.0001) risk stratification among low vs. high risk groups. Beyond comparing within single-omic methods, CMVAE outperforms six state-of- the-art multi-omics survival baselines by a minimum of 3%. Availability: All codes including data preparation scripts, training and validation recipes, and experimental configurations are available at: https://github.com/kabir-ai2bio-lab/CMVAE. Note: Accepted at the IEEE BIBM 2026 workshop RLPD-Bio (Representation Learning and Pattern Discovery for Bioinformatics and Biomedicine).
Related
- Synthetic MPRA perturbation supervision improves variant-effect prediction in sequence-to-function models
- KLHL40 gene replacement therapy in severe nemaline myopathy improves survival and skeletal muscle function in a preclinical mouse model
- Hepatocellular carcinoma screening in patients with compensated hepatitis C virus (HCV)-related cirrhosis aware of their HCV status improves survival: A modeling approach
- Type 1 immunity linked to CMV exposure improves long-term survival in sepsis without shock
- Bacterial toxin sensing induces transient inflammasome-dependent innate immune memory that improves survival during sepsis
BibTeX
@misc{singh2026cmvae,
title = {{CMVAE: Contrastive Supervision Improves Single-Omic Latent Manifold for Breast Cancer Survival Prognosis}},
author = {A. Singh and A. Kohut and J. M. Templeton and A. Kabir},
howpublished = {bioRxiv},
year = {2026},
month = oct,
doi = {10.64898/2026.10.30.763085},
url = {https://www.biorxiv.org/content/10.64898/2026.10.30.763085v1},
note = {CC BY},
}
RIS
TY - UNPB
TI - CMVAE: Contrastive Supervision Improves Single-Omic Latent Manifold for Breast Cancer Survival Prognosis
AU - A. Singh
AU - A. Kohut
AU - J. M. Templeton
AU - A. Kabir
PY - 2026
DA - 2026/10/30
PB - bioRxiv
DO - 10.64898/2026.10.30.763085
UR - https://www.biorxiv.org/content/10.64898/2026.10.30.763085v1
AB - High-dimensional omics data are vital for fore- casting patient prognosis in oncology. While deep multi-omics learning methods are increasingly popular, they suffer from data missingness and low patient overlap across modalities. Consequently, advancing single-omic representation learning re- mains a critical yet challenging research problem. Standard generative dimensionality reduction methods rely on recon- struction objectives that optimize the global manifold topology; however, overlook downstream clinical outcomes. To address these limitations, we introduce a Contrastive Masked Denois- ing Variational Autoencoder (CMVAE), a self-supervised dual- optimization framework designed to map single-omic information into a clinical outcome-aware latent space. A stochastic feature masking strategy corrupts the input omic profiles, and CMVAE is optimized to simultaneously infer missing molecular data and contrastive alignment of patient survival trajectories. We evaluate CMVAE using leak-proof five-fold stratified cross-validation on the TCGA Breast Invasive Carcinoma (BRCA) cohort. The method outperforms traditional unsupervised methods, achieving an AUROC of 0.7397 {+/-} 0.0239 on patient prognosis. Further analysis demonstrates a highly significant (p < 0.0001) risk stratification among low vs. high risk groups. Beyond comparing within single-omic methods, CMVAE outperforms six state-of- the-art multi-omics survival baselines by a minimum of 3%. Availability: All codes including data preparation scripts, training and validation recipes, and experimental configurations are available at: https://github.com/kabir-ai2bio-lab/CMVAE. Note: Accepted at the IEEE BIBM 2026 workshop RLPD-Bio (Representation Learning and Pattern Discovery for Bioinformatics and Biomedicine).
ER -