Key Points
- Scientists have used a supercomputer to create a new AI model that understands the “language of proteins.”
- The model, called PLM-Interact, can predict protein-protein interactions.
- This can help us understand how diseases like cancer and viral infections develop.
- The new AI is significantly more accurate than other state-of-the-art models.
Scientists have repurposed a powerful supercomputer, normally used to study the mysteries of the universe, to train a new AI model that can translate the “language of proteins.” The breakthrough published in Nature Communications could give us a much better understanding of how diseases like cancer develop and how viruses infect our cells.
The new large language model, called PLM-Interact, was developed by a team at the University of Glasgow. It is designed to understand how the thousands of different proteins in our bodies interact with one another, a complex and vital process at the heart of all life. When these interactions go wrong, it can lead to disease.
Until now, studying these protein interactions has been slow and expensive. This new AI tool can predict how proteins will “talk” to each other with unprecedented accuracy, and even which specific genetic mutations will cause those conversations to go awry.
To build the model, the researchers trained it on a massive dataset of over 421,000 human protein pairs, using a supercomputer that was originally designed to simulate the workings of the universe.
The results are impressive. Early tests show that PLM-Interact is significantly more accurate than other leading AI protein models, including Google DeepMind’s AlphaFold3. It also accurately predicted how viruses interact with human proteins, a tool that could be invaluable in fighting future pandemics.
