Scientists have developed a groundbreaking physics-based algorithm aimed at designing biomolecules with custom properties, addressing some of the limitations of the renowned AI tool AlphaFold. Jointly led by researchers Ryan Krueger and Krishna Shrinivas, this project was a first for both. It is a huge advance in the ability to study intrinsically disordered proteins, which have been difficult for existing AI approaches to generate accurate predictions for.
AlphaFold, by Google DeepMind, has revolutionized protein prediction and was recognized by a recent Nobel Prize for that work. As a free and open AI-based tool, this new P@IS offers remarkable capabilities for the prediction of protein structures. It runs into significant roadblocks with almost 30% of the proteins encoded by the human genome. The similar limitations of AlphaFold emphasize the critical need for new approaches. Such progresses demand new haronies that can provide more profound aspects into such complex biomolecules.
One such recent project, published in Nature Computational Science, seeks to address these limitations of AlphaFold. More broadly, it showcases the amazing importance of intrinsic disorder in proteins. Shrinivas got interested in this space due to how much it challenges existing AI-based approaches. He stresses the need to keep pushing the envelope on techniques to better understand these hard-to-grasp proteins.
Advancements in Protein Prediction
A big leap forward came with Google’s AlphaFold, which used cutting-edge machine learning to transform the discipline of protein structure prediction. It has clear limitations, particularly with intrinsically disordered proteins, which lack a stable structure and are challenging to model accurately. These misfolding proteins, such as alpha-synuclein, are strongly associated with several diseases such as Parkinson’s disease.
Ryan Krueger and Krishna Shrinivas saw these gaps and wanted to create a better way. They carried forward a physics-based algorithm, frequently used to optimize molecular dynamics simulations. Their focus was to improve the design and prediction of biomolecules with defined, separate, biologically relevant conformations.
Michael Brenner is the Gordon McKay Professor of Applied Mathematics and Applied Physics at the School of Engineering and Applied Sciences (SEAS). It was these early experiments that allowed Gus to see the broader potential of this quickly developed technique. His lab’s expertise in computational methods informed the larger picture that helped shape the development of this new algorithm.
The Role of Disordered Proteins
Historically, intrinsically disordered proteins have been difficult to work with for many researchers, but they have become an area of great interest for continued research. Shrinivas’s fascination with these proteins comes from how they behave, the intricate roles that they play in biological systems. Unfortunately, traditional AI models such as AlphaFold are not able to account for the complexities involved with disordered regions. As such, they frequently create flawed or erroneous projections.
A newly developed physics-based algorithm aims to address this specific gap. It provides a richer conceptual basis for explaining how and why these proteins exert their activity across multiple biological contexts. Rigorous computational design of biomolecules with novel properties and functions would unlock powerful advances in fundamental science. Alongside this presents a fantastic opportunity to create new therapeutic applications.
The research team’s work is an example of how interdisciplinary approaches can be key to answering complex biological questions. They’re applying fundamentals from physics and state-of-the-art computational approaches. This field-based, human-centric approach is designed to lead to breakthroughs that radically outperform what today’s AI tools are capable of.
Implications for Future Research
The results of this project hold important promise for expanding research in molecular biology and biophysics. Scientists are looking to understand proteins on a much deeper, functional level. Getting to know these IUPs inside and out will be instrumental to creating novel therapies and drug inventions.
Their research, recorded at https://doi.org/10.1038/s43588-025-00881-y explores frequent examples of how physics-based approaches can enhance or be integrated with popular AI techniques such as AlphaFold. This integration of AI with MD offers the potential to build more detailed models that capture the full complexity of protein structures and their interactions.