Extended Data Fig. 8: Multi-task DEBIAS-M improves classification and regression of metabolite levels. | Nature Microbiology

Extended Data Fig. 8: Multi-task DEBIAS-M improves classification and regression of metabolite levels.

From: Processing-bias correction with DEBIAS-M improves cross-study generalization of microbiome-based prediction models

Extended Data Fig. 8

a, Same as Fig. 5c, showing 399 metabolites with >50 observations for which MelonnPan51 does not provide a prediction, using samples from 218 individuals. b, Same as (a), but classifying samples where a metabolite had levels that were higher than the median, for all metabolites. This analysis was performed for 120 metabolites selected by MelonnPan out of 509 metabolites that had values lesser and greater than their median in both batches, such that a classification task could be used to train and evaluate a model. Samples from 218 individuals were used. The multi-task version of DEBIAS-M outperformed both its single-task version (one-sided Wilcoxon signed-rank p = 0.02, as well as MelonnPan (p = 5.6 × 10−14). Box, IQR; line, median; whiskers, nearest point to 1.5*IQR.

Back to article page