Predicting lasso peptide structure with LassoPred

Lasso peptides (LaPs) are a structurally unique class of ribosomally synthesized and post-translationally modified peptides (RiPPs), characterized by a threaded rotaxane topology that confers exceptional stability and a broad spectrum of bioactivities. Despite increasing interest in LaPs as antimicrobial and therapeutic agents, accurate structure prediction remains a major challenge due to the scarcity of homologous templates and the failure of mainstream prediction tools to capture their knotted topology. In this chapter, we present LassoPred, a modular, machine learning-guided computational pipeline for high-throughput 3D structure prediction of lasso peptides from sequence alone. LassoPred integrates support vector machine classifiers trained on ESM2 embeddings to annotate critical topological features-namely, the isopeptide ring and plug residues-and a topology-aware structure constructor that assembles and refines atomic models via homology modeling, residue mutation, and energy minimization. The tool is implemented in Python, compatible with AMBER and PyMOL, and accessible via a public web interface, enabling both expert and non-specialist users to submit sequences and retrieve optimized structural models. LassoPred demonstrates near-experimental accuracy while reducing prediction time to minutes, expanding known lasso peptide structural coverage from fewer than 50 experimentally determined structures to over 4000 genome-mined models. The pipeline is further extensible to engineered LaP variants and potentially other RiPP families. By bridging the gap between sequence discovery and structural insight, LassoPred facilitates downstream applications in enzymology, structural biology, synthetic biology, and therapeutic design.