# Multi-View Variational Autoencoder for Missing Value Imputation in Untargeted Metabolomics

**Authors:** Chen Zhao, Kuan-Jui Su, Chong Wu, Xuewei Cao, Qiuying Sha, Wu Li, Zhe Luo, Tian Qin, Chuan Qiu, Lan Juan Zhao, Anqi Liu, Lindong Jiang, Xiao Zhang, Hui Shen, Weihua Zhou, Hong-Wen Deng  
**arXiv:** [2310.07990](https://arxiv.org/abs/2310.07990)  
**Date:** 2023-10-12  
**Summary page:** https://www.summarizepaper.com/en/arxiv-id/2310.07990v2/

## Key points

- Missing data is a common challenge in mass spectrometry-based metabolomics, leading to biased and incomplete analyses.
- Integration of whole-genome sequencing (WGS) data with metabolomics data enhances accuracy of data imputation.
- A novel method utilizing a multi-view variational autoencoder was proposed to impute unknown metabolites based on genomic information.
- The study enrolled 1,110 subjects with WGS and metabolomics data to investigate genetic and environmental risk factors for osteoporosis and musculoskeletal diseases.
- Evaluation of the method demonstrated its superiority over conventional imputation techniques, achieving R^2-scores > 0.01 for 71.55% of metabolites.
- Integration of WGS data enhances data completeness and improves downstream analyses in precision medicine research, offering valuable insights into metabolic pathways and disease associations.

## Summary

, , , , 

Background: Missing data is a common challenge in mass spectrometry-based metabolomics, which can lead to biased and incomplete analyses. The integration of whole-genome sequencing (WGS) data with metabolomics data has emerged as a promising approach to enhance the accuracy of data imputation in metabolomics studies. Previous research has highlighted the importance of multi-omics data integration for identifying biomarkers and understanding disease mechanisms [17]. Leveraging information from WGS data and reference metabolites, a novel method utilizing a multi-view variational autoencoder was proposed to impute unknown metabolites based on genomic information. Method: To investigate genetic and environmental risk factors for osteoporosis and musculoskeletal diseases, the study enrolled 1,110 subjects from the Louisiana Osteoporosis Study (LOS) who had both WGS and metabolomics data [16]. The detailed procedure for WGS involved sequencing human peripheral blood DNA with an average read depth using a BGISEQ-500 sequencer. For feature extraction and missing metabolomics data imputation, the proposed method utilized burden score, polygenic risk score (PGS), and linkage disequilibrium (LD) pruned single nucleotide polymorphisms (SNPs). Figure 1 illustrates the architecture of the multi-view variational autoencoder used for this purpose. Results: Evaluation of the method on empirical metabolomics datasets with missing values demonstrated its superiority over conventional imputation techniques. By incorporating 35 template metabolites derived from burden scores, PGS, and LD-pruned SNPs, the proposed method achieved R^2-scores > 0.01 for 71.55% of metabolites. This performance improvement showcases the effectiveness of integrating WGS data in metabolomics imputation. Conclusion: The integration of WGS data not only enhances data completeness but also improves downstream analyses in precision medicine research. By leveraging multi-modal data integration, researchers can conduct more comprehensive and accurate investigations into metabolic pathways and disease associations. This study offers valuable insights into the potential benefits of utilizing WGS data for metabolomics imputation, emphasizing the importance of incorporating genomic information in advancing precision medicine approaches. Additionally, previous studies have emphasized the power and predictive accuracy of polygenic risk scores [23], highlighting their relevance in enhancing our understanding of complex diseases through multi-omic approaches.