Fig. 1: Development of a lasso peptide-specific language model, LassoESM.
From: LassoESM a tailored language model for enhanced lasso peptide property prediction
A Lasso peptide biosynthesis requires a leader peptidase, RiPP recognition element (RRE), and lasso cyclase to tie a linear core peptide into the lariat-like knot. B LassoESM was built upon the ESM-2 architecture and further pre-trained on lasso peptides using a domain-adaptive approach with masked language modeling. The resulting LassoESM embeddings were utilized for three downstream tasks: predicting lasso cyclase substrate tolerance, identifying substrate compatibility between non-cognate pairs of lasso cyclases and substrate peptides, and predicting RNAP inhibitory activity (numbers indicate enrichment values, estimating RNAP inhibitory activity).
