Skip to main content
. 2021 Sep 8;38(1):157–163. doi: 10.1093/bioinformatics/btab645

Fig. 2.

Fig. 2.

CoDaCoRe variable selection for the first (most explanatory) log-ratio on the Crohn disease data (Rivera-Pinto et al., 2018). For each of 10 independent bootstraps of the training set (80% of the data randomly sampled with stratification by case–control), we show which variables are selected in the numerator (blue) and denominator (orange) of the balance. CoDaCoRe learns remarkably consistent log-ratios across independent training sets