Table 3.

Unsupervised machine learning methods used in combination with constraint-based models.

Unsupervised ML method	Description	Applications
PCA	Dimensionality reduction Data interpretability Data simplification Identification of variation sources	Dimensionality reduction of fluxomic data [85], [86] Can be applied to CBM results to identify central fluxes and pathways [85,86,95,96) Identification of metabolism active EMs [87], [88], [89], [90] Can be applied to experimental or simulation data to further inform downstream CBMs or ML algorithms [91], [92], [93], [94]
Clustering	Identification of sub-populations Clustering criteria, such as centroid- or distribution-based, vary depending on the chosen algorithm	Hierarchical clustering for the identification of populations and the development of population specific CBMs [98], [100] Gene network reconstruction [97] Heatmaps of metabolomics data for studying metabolic alterations [99] k-means clustering to group CBM results [100]
Autoencoders	Unsupervised artificial neural networks Dimensionality reduction	Variational autoencoders (VAEs) in combination with CBMs to identify biologically relevant features from microarray data [102] Autoencoders customized based on GeM models [104]
Bayesian factor model	Dimensionality reduction Identification of certain latent variables that account for data variation	Metabolic pathways analysis from gene expression data [105]
MCR-ALS	Dimensionality reduction Application of custom constraints	Pathways identification [106]