← Latest papers
📄 molecular biology

Predicting conformational changes induced by biomolecular interactions

The paper introduces K-Fold, a novel generative learning framework that predicts biomolecular conformational changes by modeling the explicit structural transition from unbound to bound states via an ultrafast diffusion bridge, thereby overcoming the limitations of current co-folding models in capturing interaction-induced dynamics without relying on multiple sequence alignments.

Original authors: Bae, Y., Baek, J., Choi, E., Choi, J. H., Han, H., Jang, H., Jeon, W., Jo, J., Kim, D., Kim, H., Kim, J. H., Kim, J. H., Kim, S., Kim, T., Kim, Y., Lee, J., Lee, J., Moon, S., Park, M., Samuel, C., Se
Published 2026-11-19 (v2)
📖 5 min read🧠 Deep dive

Original authors: Bae, Y., Baek, J., Choi, E., Choi, J. H., Han, H., Jang, H., Jeon, W., Jo, J., Kim, D., Kim, H., Kim, J. H., Kim, J. H., Kim, S., Kim, T., Kim, Y., Lee, J., Lee, J., Moon, S., Park, M., Samuel, C., Seo, H., Seo, J., Seo, S., Shin, M., Shin, Y., Zhung, W., Ahn, S., Han, D., Hwang, S. J., Kim, H. M., Kim, W. Y., Lee, G. R., Oh, B.-H.

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 ⚕️ This is an AI-generated explanation of a preprint that has not been peer-reviewed.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

Cells are bustling cities where molecules constantly meet, shake hands, and change shape to send signals or build new structures. To understand how life works, or how to design medicines that fit perfectly into these molecular machines, scientists need to know the exact three-dimensional shape of these interacting parts. For decades, the challenge has been predicting how two separate molecules will look when they come together. While modern computers have become incredibly good at guessing the shape of a single molecule based on its genetic code, they often stumble when asked to predict how that molecule will twist and turn once it meets a partner. The problem is that molecules are not rigid blocks; they are flexible. When a drug binds to a protein, or an antibody grabs an antigen, both often shift their form to lock together. Current methods frequently miss these subtle, crucial movements, producing a shape that looks plausible but is functionally wrong.

A team of researchers has introduced a new approach called K-Fold that changes how this prediction happens. Instead of trying to guess the final, locked-together shape from scratch, K-Fold starts with the two molecules already in their separate, unbound forms. It then uses a mathematical process to simulate the journey of them coming together. Imagine a bridge that connects two distinct landscapes; this tool builds a bridge between the shape of the molecule alone and the shape of the molecule when it is bound to its partner. By treating the assembly as a transition from one state to another, rather than a single static guess, the system can better capture the specific movements required for the molecules to fit. This method does not rely on looking for evolutionary clues in genetic history, which often fail for engineered molecules or new drug targets, but instead focuses on the physical rules of how shapes adapt to one another.

The researchers tested this new framework against a wide range of biological puzzles, including how antibodies find their targets, how small drug molecules recruit proteins to break them down, and how receptors in the brain switch between active and inactive states. In tests involving antibodies, which are notoriously difficult to predict because their binding loops change rapidly, K-Fold succeeded in finding the correct binding site in more than half of the cases, outperforming previous top-tier models that relied on evolutionary data. This was particularly true for new targets where no historical examples existed in scientific databases. In the case of molecular glues—tiny compounds that force two proteins to stick together—the new model correctly assembled the three-part complex in nearly 44% of attempts, a significant improvement over existing tools that often misplaced the recruited protein entirely.

The study also looked at complex signaling proteins like kinases and G-protein coupled receptors, which act as switches in the body. These proteins have different shapes when they are turned on versus when they are turned off. Older models tended to default to the most common "on" shape, even when the conditions suggested the protein should be "off." K-Fold, however, successfully predicted the correct inactive state in about 78% of cases where the protein was bound to an inhibitor, accurately capturing the specific conformational changes that silence the signal. This ability to distinguish between different functional states is vital for drug design, as a medicine might need to lock a protein in a specific shape to stop a disease process.

Perhaps the most powerful aspect of this work is its ability to explore multiple possibilities without needing to be retrained. Because the system starts with the separate components, researchers can feed it different starting shapes for the unbound molecule to see how the final complex might look. In difficult cases where standard models failed completely, using a diverse set of starting shapes allowed the new system to find the correct binding mode. For example, in a challenging protein target, the model found the right structure in nearly 40% of attempts when guided by a diverse set of starting shapes, whereas it found nothing when using a single, standard starting shape. This suggests the system has learned the general physical principles of how molecules adapt, rather than just memorizing specific examples from its training data.

While the method is not perfect and still faces challenges in predicting the exact stability of different shapes or handling extremely rare molecular movements, it represents a significant shift in how scientists approach molecular assembly. By focusing on the transition between states rather than just the final destination, K-Fold offers a more robust way to predict how the machinery of life fits together. This approach provides a clearer path for designing new drugs and understanding cellular regulation, moving the field closer to a dynamic view of biology where molecules are seen not as static statues, but as flexible partners constantly reshaping themselves to do their work.

Drowning in papers in your field?

Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.

Try Digest →