Fig. 5

Predicted probabilities from the Bayesian regression models for training and validation datasets. Predicted probabilities for the CpG-related NEC and Gram-negative-related NEC models are shown for 11 control infants, 11 NEC cases and for the validation dataset. Dashed lines indicate the 90% quantile of predicted probability for the eleven control samples along each axis. “NEC” samples are the closest available samples to an infant’s NEC diagnosis for each dataset and “Pre-NEC” refers to any prior samples from these infants. NEC cases that received no antibiotics prior to samples being taken are marked with stars. The points for samples “N1” and “N8” are overlaid