C-LipoVAE: A Leakage-Safe Conditional Variational Autoencoder for Serum Lipidomic Phenotype Stratification
- 1. School of Chinese Medicine, The University of Hong Kong
Description
C-LipoVAE is a leakage-safe conditional variational autoencoder (CVAE) framework for binary phenotype stratification of high-dimensional omics data. It addresses label leakage in latent-space evaluation with a null-condition (zero-vector) encoding protocol, and provides C2ST AUC, MMD², symmetrised KL, active-units, and integrated-gradient attribution metrics.
This archive corresponds to the v0.1.0 release accompanying the manuscript submission to the IEEE Journal of Biomedical and Health Informatics. The reference application is serum lipidomic stratification of a phlegm-dampness Chinese-medicine body-constitution phenotype; framework generalisability is demonstrated on PIVUS/MTBLS90 lipidomics and the Wisconsin Breast Cancer benchmark. All results are real, bit-exactly reproducible pipeline outputs.
Files
Files
(69.8 kB)
| Name | Size | |
|---|---|---|
|
md5:c55ed621c6e0aafa6dbcdc8339e578d5
|
69.8 kB | Download |
Additional details
Related works
- Is derived from
- https://github.com/23008613g/C-LipoVAE (URL)