Proteins are the building blocks of life, regulating a vast array of cellular functions in all living organisms. Their three-dimensional structure is crucial to their function, but determining this structure has traditionally been a laborious and expensive process.

In 2020, the AI tool AlphaFold, developed by DeepMind, marked a breakthrough in protein structure prediction. Based on neural networks, AlphaFold demonstrated remarkable accuracy in predicting how proteins fold. This achievement was recognized with the 2024 Nobel Prize in Chemistry awarded to Demis Hassabis (a man we will thank for a long time!).

Despite its success, AlphaFold had limitations, including the inability to predict very large protein compounds or use experimental or incomplete data. Researchers at Linköping University have addressed these shortcomings by developing AF_unmasked, an enhanced version of AlphaFold.

AF_unmasked: Expanding the Capabilities of AlphaFold

AF_unmasked incorporates experimental data and partial information, enabling it to predict larger and more complex protein structures. It provides researchers with a more comprehensive understanding of protein structure and function.

Claudio Mirabello, docent at Linköping University, explains that AF_unmasked combines experimental data with neural networks to generate a complete picture of protein structures. It can also utilize partial structural data to produce accurate predictions. This advancement facilitates the design of new proteins, particularly for medical applications.

The Role of Collaboration and Data in AlphaFold’s Success

The development of AlphaFold was made possible by the collective efforts of researchers worldwide who have been collecting data on protein structures since the 1970s. This data formed the basis for training AlphaFold’s neural networks. Additionally, the availability of supercomputers equipped with GPUs was crucial for performing the complex calculations required by AlphaFold.

But let’s not forget the news AI tool that DeepMind made available to researchers, AlphaProteo, which can design entire proteins on demand…there will soon be exciting news in this field applied to healthcare!

Data source