Skip to main navigation Skip to search Skip to main content

GCVAE: A graph conditional variational autoencoder for feature dependency guided synthetic metabolomic data generation

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution

Abstract

Metabolomics benefits biomarker discovery and other applications, but publicly available metabolomic data is very limited. Synthetic metabolomic data can aid novel research paradigms. Due to curse of dimensionality, real metabolomic data are often incompatible with major generative approaches such as conditional variational autoencoder (CVAE) and conditional generative adversarial network (CGAN). We present a novel graph conditional variational autoencoder (GCVAE) to exploit feature dependency graphs - capturing relationships in real data - to restrict the exponential space of feature combinations by aligning learned distributions with the graph, which benefits small training data. We evaluate GCVAE for synthetic metabolomic data generation from 4 aspects - fidelity, correlation, utility, and privacy. We show that GCVAE often generates higher quality synthetic data than CVAE and/or CGAN. We show that GCVAE can further improve synthetic data by utilizing domain knowledge on metabolic reactions. Overall, our results show that the feature dependency-guided GCVAE is a promising approach for generating synthetic metabolomic data from limited real data.

Original languageEnglish (US)
Title of host publicationBCB 2025 - Proceedings of the 16th ACM International Conference on Bioinformatics, Computational Biology, and Health Informatics
PublisherAssociation for Computing Machinery
ISBN (Electronic)9798400722004
DOIs
StatePublished - Dec 10 2025
Event16th ACM Conference on Bioinformatics, Computational Biology, and Health Informatics, ACM-BCB 2025 - Philadelphia, United States
Duration: Oct 12 2025 → Oct 15 2025

Publication series

NameBCB 2025 - Proceedings of the 16th ACM International Conference on Bioinformatics, Computational Biology, and Health Informatics

Conference

Conference16th ACM Conference on Bioinformatics, Computational Biology, and Health Informatics, ACM-BCB 2025
Country/TerritoryUnited States
CityPhiladelphia
Period10/12/25 → 10/15/25

Keywords

  • correlation graph
  • domain knowledge
  • graph conditional variational autoencoder
  • knowledge graph
  • metabolomics
  • synthetic data

ASJC Scopus subject areas

  • Computer Science Applications
  • Software
  • Biomedical Engineering
  • Health Informatics

Fingerprint

Dive into the research topics of 'GCVAE: A graph conditional variational autoencoder for feature dependency guided synthetic metabolomic data generation'. Together they form a unique fingerprint.

Cite this