Researchers at Linköping University have made a significant breakthrough in protein structure prediction by improving the AI tool AlphaFold to predict very large and complex protein structures. This achievement, published in Nature Communications, is a crucial step towards more efficient development of new proteins for medical drugs. Proteins are essential molecules that regulate cell functions, and their shape determines their function. For over 50 years, researchers have been trying to predict and design protein structures, but it has been a laborious and expensive task.
In 2020, Deepmind released AlphaFold, an artificial intelligence based on neural networks that can predict protein folding with great accuracy. However, the programme had limitations, including inability to predict very large protein compounds or draw conclusions from experimental data. Claudio Mirabello and Björn Wallner, researchers at Linköping University, have now developed AlphaFold further to overcome these shortcomings, creating a new tool called AF_unmasked that can take in information from experiments and partial data as well as predict very large and complex protein structures. This breakthrough has the potential to revolutionize protein design and development of medical drugs.
Predicting Protein Structures with AlphaFold: A Breakthrough in AI-Assisted Biology
The development of AlphaFold, an artificial intelligence (AI) tool, has revolutionized the field of protein structure prediction. Researchers at Linköping University have taken this technology a step further by improving its capabilities to predict very large and complex protein structures, as well as integrating experimental data into the tool.
Proteins are essential molecules in living organisms, consisting of 20 different amino acids that stick together in long rows. The sequence and structure of these proteins determine their functions, which can range from controlling muscles and forming hair to transporting oxygen into the blood and digesting food. With billions of possible combinations, predicting protein structures has been a laborious and expensive task involving manual handling.
The release of AlphaFold in 2020 marked a breakthrough in AI-assisted biology, allowing researchers to predict protein structures with great accuracy. This innovation was recognized with the Nobel Prize in Chemistry in 2024. However, the initial version of AlphaFold had limitations, including its inability to predict very large protein compounds and draw conclusions from experimental or incomplete data.
Overcoming Limitations: AF_unmasked
Researchers at Linköping University have addressed these shortcomings by developing AF_unmasked, an improved version of AlphaFold. This tool can now take in information from experiments and partial data, as well as predict very large and complex protein structures. By providing guidance on how to design proteins based on experimental results, AF_unmasked enables researchers to refine their experiments and build larger structures.
The integration of experimental data into AF_unmasked is a significant advancement, allowing researchers to feed in draft structures and obtain relatively accurate results. This development has the potential to accelerate the discovery of new protein drugs and deepen our understanding of protein functions.
The Power of Collaboration: From Database to Breakthrough
The AlphaFold breakthrough was made possible by the collective efforts of researchers worldwide who have been collecting data on protein structures since the 1970s. This database provided the training data necessary for AlphaFold’s development. The technological advancement of supercomputers using graphics processing units (GPUs) for heavy calculations also played a crucial role in enabling large-scale predictions.
Björn Wallner, professor of bioinformatics at Linköping University, has worked with one of the Nobel Prize winners and emphasizes the endless possibilities for protein design. “You always have to find new, more difficult problems when you have solved the old ones,” he says.
Building on AlphaFold: A Precursor and a Legacy
Claudio Mirabello, docent at Linköping University, and Björn Wallner developed a precursor to AlphaFold that inspired Deepmind in developing the tool. The resources of the Google-owned company enabled the development of what is now an indispensable tool for protein scientists worldwide.
Mirabello highlights the significance of their earlier work, stating that “AlphaFold wasn’t the first tool to use deep neural networks to solve the problem… So, you could say that AlphaFold was based on our idea, and now we are building on AlphaFold.” This legacy underscores the importance of collaboration and knowledge sharing in driving scientific progress.
The advancements made by researchers at Linköping University demonstrate the potential for AI-assisted biology to revolutionize our understanding of protein structures and functions. As the possibilities for protein design continue to expand, it is clear that this field will remain a vibrant area of research driven by innovation and collaboration.
External Link: Click Here For More
