AlphaFold3 Can't Do It, But It Can! Shanghai Jiao Tong & Vanderbilt Develop LassoPred: A Tool for Predicting Lasso Peptide 3D Structures}
Researchers from Shanghai Jiao Tong and Vanderbilt created LassoPred, a tool to predict the 3D structures of lasso peptides, expanding structural databases and aiding drug discovery.
Editor | Carrot Skin
Lasso peptides (LaP), characterized by their knotted, active loop structures, are a class of ribosomally synthesized and post-translationally modified peptides (RiPPs). They can serve as antibiotics, enzyme inhibitors, and molecular switches.
Although bioinformatics predicted thousands of LaP sequences, only about 50 have been characterized structurally over the past 30 years.
Existing computational tools like AlphaFold2, AlphaFold3, and ESMfold struggle to accurately predict LaP structures due to their irregular knotted frameworks and the presence of isopeptide bonds.
To address this challenge, researchers from Shanghai Jiao Tong University and Vanderbilt University developed LassoPred, which includes a classifier to annotate LaP sequences with loops, rings, and tails, and a builder to construct 3D structures.
Using LassoPred, the team predicted the structures of 4,749 unique core LaP sequences, creating the largest computational database of lasso peptide structures to date.
This study, titled LassoPred: a tool to predict the 3D structure of lasso peptides, was published on July 1, 2025, in Nature Communications.
Link to the paper: https://www.nature.com/articles/s41467-025-60544-4
LassoPred's Design and Applications
LassoPred is accessible via web interface and command-line tools, supporting future structure-function relationship studies and helping discover functional lasso peptides for chemical and biomedical applications.
Website: https://lassopred.accre.vanderbilt.edu/
Diagram: Architecture of LassoPred (Source: Paper)
The LassoPred database and tool are expected to deepen understanding of LaP, accelerate the discovery of new functional peptides, and aid in the design and engineering of lasso peptides. The database has expanded from 47 to 4,749 structures.
Researchers note that the training dataset contains only 47 experimentally characterized LaPs, reflecting the scarcity of known structures, most of which are determined by NMR, with fewer by X-ray crystallography. Structural determination often takes months or years, with no guarantee of success.
To validate LassoPred, researchers attempted to determine the structure of a new antimicrobial LaP via NMR. Despite good expression in heterologous hosts, changing solvents, temperature, and NOE mixing times failed to produce high-quality data for structure elucidation.
This challenge highlights the importance of LassoPred, which can provide large-scale computational structure predictions beyond known structures, useful for docking simulations and drug discovery.
The predicted structures help elucidate LaP’s remarkable thermal stability and solvent stability, inspiring engineering strategies to modify properties. Constructing LaPs with non-natural knotted folds can also shed light on folding landscapes and chirality origins, enriching fundamental knowledge.
Illustration: Database and prediction tools for lasso peptides (Source: Paper)
LassoPred can help prioritize new LaP discoveries, such as antibiotics and self-assembling biomaterials. Its database includes thousands of sequences and structures for machine learning models like DeepLasso, facilitating de novo design of lasso peptides.
Paper link: https://www.nature.com/articles/s41467-025-60544-4