Lasso peptides, a class of naturally occurring bacterial compounds, have emerged as promising candidates for next-generation therapeutics due to their unique slipknot structure, which confers exceptional stability and diverse bioactivities, including antibacterial, antiviral, and anticancer effects. Despite their potential, the rational engineering of these molecules has been hindered by the complexity of their biosynthesis and the lack of specialized computational tools. To address this, researchers from the Carl R. Woese Institute for Genomic Biology—led by Doug Mitchell of Vanderbilt University and Diwakar Shukla of the University of Illinois Urbana-Champaign—developed LassoESM, a specialized large language model tailored to lasso peptides. Built on the ESM-2 architecture, LassoESM was further pre-trained using a domain-adaptive masked language modeling approach on a curated dataset of thousands of experimentally validated lasso peptide sequences, many of which were identified through bioinformatics and manually verified. This training enabled the model to capture the distinctive sequence-structure relationships critical to lasso formation, particularly the interaction between the core peptide and lasso cyclase enzymes, which catalyze the formation of the characteristic knot. The resulting embeddings from LassoESM were successfully applied to three key tasks: predicting substrate tolerance of lasso cyclases, identifying functional compatibility between non-cognate cyclase-peptide pairs, and forecasting RNAP inhibitory activity—demonstrating high accuracy even with limited experimental data. The model’s ability to predict enzyme-substrate compatibility is especially transformative, as it allows researchers to engineer lasso peptides with desired properties without relying on laborious trial-and-error experimentation. Unlike general-purpose protein language models such as AlphaFold, which fail to account for the unique structural constraints of lasso peptides, LassoESM provides a targeted solution that bridges the gap between sequence and function. The interdisciplinary collaboration, supported by advanced computing resources and the MMG theme at the Carl R. Woese Institute, underscores the power of integrating computational and experimental biology. The success of LassoESM not only accelerates the discovery and design of lasso-based therapeutics but also sets a precedent for developing similar specialized AI tools for other complex natural product classes. Industry experts highlight the model’s potential to streamline drug discovery pipelines, reduce development time, and enable the creation of stable, orally bioavailable peptide drugs. With future plans to expand the model to other peptide families and enhance its predictive capabilities, LassoESM marks a significant leap forward in peptide engineering and AI-driven synthetic biology.