Fig. 4

A Gene cluster family network of the 11,572 predicted RiPPs BGCs from 33,904 marine prokaryotic genomes, which consists of 4619 GCFs (3268 singletons not shown). The top ten dominant subcategories of RiPPs GCFs were labeled with different colors. B Venn diagrams of RiPPs GCF overlap in marine prokaryotic genomes of different categories. The number of GCFs and singletons (shown in parentheses) was also compared. C The prediction scores of the 853 core peptides in three models (attention, LSTM, and BERT). D GCFs containing BGCs with the core peptides can be predicted by antiSMASH. BGCs encoding core peptides predicted as AMPs were marked. Node color reflects the subcategory of RiPPs BGCs