HighFold-Lasso: Topology-Guided Prediction of Lasso Peptide Structures Using AlphaFold3
HighFold-Lasso is a novel computational framework that integrates custom covalent connectivity with scaffold-guided adaptive initialization to achieve 100% success in predicting the complex mechanically interlocked structures of all four lasso peptide classes, significantly outperforming general-purpose models in accuracy and reliability.
Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of a preprint that has not been peer-reviewed. It is not medical advice. Do not make health decisions based on this content. Read full disclaimer
Nature has a knack for building molecular knots that seem impossible to untie. Among the most intricate of these are lasso peptides, a special family of proteins synthesized by bacteria and other microbes. Imagine a tiny loop of string with a tail that has been pulled through the center of the loop and then locked in place, so the tail cannot slip back out. This is the lasso peptide: a ring formed by a chemical bond, with a tail threaded through it and held fast by a second bond or a physical bulge. This mechanical interlock gives the molecule remarkable strength, allowing it to withstand high heat and resist being broken down by enzymes. Because of this durability, scientists are eager to use these peptides as building blocks for new medicines and industrial tools. However, to engineer them effectively, researchers first need to know exactly what they look like in three dimensions.
For years, determining the shape of these knotted molecules has been a stubborn problem. While scientists can easily read the genetic code that tells a cell how to build the peptide, translating that code into a 3D shape is notoriously difficult for computer models. The standard tools used to predict protein structures, which have become incredibly powerful in recent years, often fail with lasso peptides. These general models are excellent at guessing shapes for typical proteins, but they struggle with the specific mechanical constraint of a tail threaded through a ring. Without the right starting point, the computer might predict a shape that looks chemically correct but lacks the crucial knot, or it might simply give up. This gap between knowing the sequence and seeing the shape has left a large portion of the potential of lasso peptides unexplored.
To bridge this gap, a team of researchers at Macao Polytechnic University developed a new method called HighFold-Lasso. Instead of asking a computer to guess the shape from scratch, the team designed a system that starts with a known, valid lasso structure and gently refines it. They realized that the hardest part of the problem is not the chemistry of the bonds, but the topology—the fact that the tail must pass through the ring. By providing the computer with a "scaffold," or a rough template that already has the tail threaded correctly, the model can focus on getting the fine details right rather than trying to invent the knot itself. This approach combines the power of a modern structure-prediction engine with a specific guide that ensures the mechanical lock is never lost during the calculation.
The researchers tested their new tool on a set of 52 lasso peptides whose shapes had already been confirmed by experiments. In every single case, HighFold-Lasso successfully predicted a structure that maintained the correct threaded topology. This was a perfect score, whereas the best existing general-purpose models only managed to get the knot right about two-thirds of the time. Beyond just getting the knot right, the new method produced shapes that were much closer to the experimental reality. When the researchers measured the distance between the predicted atoms and the actual atoms, the average error was less than 2.4 angstroms for the whole molecule. This is a significant improvement over previous methods, which often had errors nearly twice as large. The team found that this precision held true even for the most complex versions of these peptides, including those with extra chemical bonds or unusual building blocks that standard models often ignore.
A key part of the success was how the team handled the starting point of the calculation. Instead of letting the computer begin with a completely random, jumbled mess of atoms, they started with a structure that already had the correct threading relationship. They then added a small amount of controlled noise to this starting shape and asked the computer to clean it up, a process that allowed the model to explore slight variations without losing the essential knot. This "low-noise" strategy proved vital. When the researchers tested what happened if they removed this specific starting guide, the success rate of predicting the correct knot dropped significantly, falling to about three-quarters of the cases. This confirmed that the initial scaffold was not just a helpful hint, but a necessary foundation for the model to work.
The study also addressed the issue of physical realism. While the computer predictions were topologically correct, the bonds between atoms were sometimes slightly stretched or compressed, which would not happen in a real, stable molecule. To fix this, the team used a physics-based refinement step that acted like a gentle massage for the structure, adjusting the bond lengths and angles to match the laws of chemistry without breaking the knot. This process removed tiny clashes where atoms were too close to each other and corrected the geometry of the special chemical links that hold the lasso together. The result was a structure that was not only topologically accurate but also physically plausible, ready to be used in further studies.
The implications of this work extend beyond just getting the numbers right. The researchers showed that their method could also handle lasso peptides that contain non-standard amino acids, the building blocks that nature sometimes swaps in to create new chemical properties. This flexibility is crucial for the future of molecular engineering, where scientists want to design custom peptides with specific functions. By proving that they can predict the shape of these complex, knotted molecules with high accuracy, the team has provided a reliable tool for exploring the vast library of lasso peptides waiting to be discovered in nature. The work suggests that for molecules with unique mechanical constraints, providing the right structural context is just as important as the chemical data itself, offering a new path forward for understanding and designing the intricate machinery of life.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.