Fig. 3: Relative importance of genetic and environmental factors for predicting the horizontal transfer of antibiotic resistance genes.

In all instances, the bars show the mean +/− SD of the importance of each factor to the accuracy of the model (MeanDecreaseAccuracy) over ten iterations. Permutation tests were used to generate a p-value for each factor and iteration. All individual p-values are available in the Source Data. *P < 0.01 across all model iterations. a The importance of the factors included in the general random forest model, based on all observed transfers (n = 1565), ordered according to their overall contribution to the accuracy of the model. b–d The mean importance of each factor group (genetic incompatibility, environmental co-occurrence, and Gram staining, respectively) for each random forest models over ten iterations. In addition to the general model (All), seven models specific to different resistance mechanisms, including aminoglycoside acetyltransferases (AAC), aminoglycoside phosphotransferases (APH), class A, C, D beta-lactamases, class B beta-lactamases, Erm 23S rRNA methyltransferases, tetracycline efflux pumps (Tet efflux), and tetracycline ribosomal protection genes (Tet RPG) are included. Signs have been added to show whether an increased value of the variable is generally indicative of horizontally spread resistance genes (+) or not (–) based on partial dependence analysis. The number of observed transfers making up the training + test data for each model is included in the legends. An equal number of randomized transfers were used for each model as the negative dataset. Source data are provided as a Source Data file.