• Scientists from the University of Sheffield and AstraZeneca have created an AI approach that could make designing proteins for new medicines easier

• This new AI approach outperforms current methods for "inverse protein folding" – a critical part of the process of creating new proteins with specific functions

• Inverse protein folding is important but difficult because small changes in protein sequences can cause unpredictable effects on the protein's structure

• For medicines to work properly, proteins need to fold into very specific 3D shapes

• This new approach works like a guide that predicts the most important folds in the protein structure, making the design process more accurate

An AI approach developed by researchers from the University of Sheffield and AstraZeneca, could make it easier to design proteins needed for new treatments.

In a study published in the journal Nature Machine Intelligence, Sheffield computer scientists in collaboration with AstraZeneca and the University of Southampton have developed a new machine learning framework that has shown the potential to be more accurate at inverse protein folding than existing state-of-the-art methods.

Inverse protein folding is a critical process for creating novel proteins. It is the process of identifying amino acid sequences, the building blocks of proteins, that fold into a desired 3D protein structure and enable the protein to perform specific functions. Protein engineering plays a critical role in drug development by designing proteins that can bind to specific targets in the body. However, the process is challenging due to the complexity of protein folding and the difficulty in predicting how amino acid sequences will interact to form functional structures.

Scientists have turned to machine learning to more accurately predict which amino acid sequences will fold into stable, functional protein structures. These models are trained on large datasets of known protein sequences and structures to improve inverse folding predictions.

The new machine learning framework, called MapDiff, from the University of Sheffield, AstraZeneca and the University of Southampton, outperformed the most state-of-the-art AI in making successful predictions in simulated tests. The results are a promising basis to develop the technology further, which, if successful, could accelerate the design of the key proteins needed to develop new vaccines and gene therapies, and other therapeutic modalities. It also complements other recent advances, such as AlphaFold, which predicts a protein’s 3D structure by reversing the approach, starting with the protein fold and retrieving the potential amino acid sequences.

Haiping Lu, Professor of Machine Learning at the University of Sheffield and the corresponding author of the study, said: “This work represents a significant step forward in using AI to design proteins with desired structures. By learning how to generate amino acid sequences that are likely to fold into specific 3D structures, our method opens new possibilities for designing new therapeutic proteins, which can be used in various therapeutic applications. It’s exciting to see AI helping us tackle such a fundamental challenge in biology.”

Peizhen Bai, Senior Machine Learning Scientist at AstraZeneca, who developed the AI as part of his PhD at the University of Sheffield’s School of Computer Science, said: “During my PhD, I was motivated by the potential of AI to accelerate biological discovery. I’m proud that our method, MapDiff, helps design protein sequences that are more likely to fold into desired 3D structures — a key step towards advancing next-generation therapeutics.”

The study was a result of a non-funded collaboration of combined industry expertise, and builds on previous work between the Sheffield computer scientists and AstraZeneca, which developed DrugBAN – AI that can predict whether a candidate drug will bind with its intended target protein molecules inside the human body and could accelerate the discovery of new medicines. The paper became one of the most cited papers from the journal Nature Machine Intelligence in 2023.

The latest paper, Mask-prior-guided denoising diffusion improves inverse protein folding, is published in Nature Machine Intelligence. To read the paper in full, visit: https://doi.org/10.1038/s42256-025-01042-6.