Extended Data Fig. 4: The sensitivity of iMAT-WPS predictions.

a, Recall of directionally constrained reactions with random subsampling of WPS data. WPS data were subsampled at different depths (i.e., number of perturbations included) for 30 times, followed by repeating iMAT-WPS using each subsample. Each dot represents the modeling result from one subsample. Y-axis shows the Jaccard index between directionally constrained reactions in full-data modeling and in a subsample analysis, giving the recall of original predictions. The box bounds the IQR divided by the median, and whiskers extend to a maximum of 1.5 × IQR beyond the box. b, An example reaction (RM04432, formula indicated) showing the distribution of OFD flux in subsampling analysis. OFD flux values are normalized to per unit bacterial uptake to facilitate cross comparisons. Each dot represents the flux of RM04432 reaction from one subsample. Any subsample flux that was within ± 30% of the original prediction was deemed ‘recalled’. The fraction of such ‘recalled’ predictions among the 30 repeats at each sampling depth is shown in the y-axis of the bottom plot. The box bounds the IQR divided by the median, and whiskers extend to a maximum of 1.5 × IQR beyond the box. c, Histogram showing the recall fraction (y-axis of the bottom plot in (b)) for all reactions at 80% subsampling depth.