Fig. 1.

Workflow for MIRTH Imputation of Metabolomics Data. a Individual datasets are normalized and rank-tranformed, accounting for left-censoring. b Preprocessed datasets (${\mathit{D}}_{\mathit{i}}$) are combined into a sparse aggregate data matrix ($\mathit{X}$), which is then factorized into embedding matrices $\mathit{W}$ and $\mathit{H}$. The product $\mathit{WH}$ yields an imputed data matrix ($\widehat{\mathit{X}}$). c Aggregate data from 9 pan-cancer metabolomics datasets with tumor and normal samples reveals poor across-dataset metabolite feature overlap and high degree of missingness