Fig. 7
From: Conditional similarity triplets enable covariate-informed representations of single-cell data
Evaluation of how covariate dependency impact accuracy in the preterm and lung cancer datasets. We examined how the extent to which each of the possible covariates and the outcome to be predicted influence outcome prediction in the preterm (PreE is used to denote the covariate of preeclampsia status) and lung cancer datasets (AE denotes adverse events). Higher values of ‘covariate dependency’ (horizontal axis) imply stronger alignment between a covariate and the outcome. CytoCoSet was run over 30 trials, under each covariate. Here, covariates are ordered as a function of increasing covariate dependency score. We report AUC (dots, vertical axis) as the metric of success to quantify how useful the embeddings were in the classification task