Fig. 1: Overview of the analysis pipeline.

First, bacterial genomes were screened for antibiotic resistance genes (ARGs). Second, phylogenetic trees were constructed using the identified protein sequences, and horizontal transfer was inferred from the trees by detecting similar genes carried by evolutionarily distant hosts. For each identified instance of horizontal transfer, data describing genetic incompatibility and co-occurrence in bacterial communities was collected. Here, genetic incompatibility was estimated as the nucleotide composition dissimilarity between the involved genomes and their ARGs, while environmental co-occurrence was estimated by mapping the involved genomes onto a large metagenomic dataset. Finally, the collected data was used to train random forest models. Factors influential for predicting horizontal transfer were identified using feature importance analysis.