Table 2.

Discriminative power of DMRs between high- and low-ranking female hyenass

Variables used for prediction *	oob error range [%]
Total rankDMRs	14.3–23.8
Selected rankDMRs	14.3–23.8
Excluded rankDMRs	42.9–57.1
Mean DNA methylation	45.1–52.4

*A random forest classifier is a learning procedure that operates by creating decision trees using a subset of the variables and a subset of the samples. The samples not used to build a tree (out-of bag-samples) are used as a test-set for the classification outcome of this tree. The average classification error on the out-of-bag-samples (oob-error) provides an estimate of the general error rate. We use oob-error as a proxy for the discriminative power of a group of rankDMRs. While oob-errors are insightful in this comparative setting, their absolute value should be taken with caution because; 1. we used the same dataset to select the rankDMR and to train the random forest classifier, 2. we did not fine-tune the hyperparameters of the model, and 3. did not investigate other learning approaches.

The table shows the range of mean classification errors on out-of-bag-samples (oob) in 1000 repetitions for different sets of variables.