An Anti-Cancer “GPS” System
The development of custom proteins for therapeutic use has just taken a giant step forward.
In 2024, half of the Nobel Prize in Chemistry went to Google DeepMind researchers John Jumper and Demis Hassabis, for their creation of AlphaFold. AlphaFold, an AI program, successfully predicted a protein’s shape and function from its chemical sequence.The other half of the prize went to David Baker of the University of Washington for solving the opposite problem: starting with the desired function of a protein, and deriving the amino acid sequence that would fold up into a molecule capable of performing that function.
And now a research team from the Technical University of Denmark has made strides toward putting this combined knowledge to use in developing a new type of immunotherapy for cancer.
The new therapy depends on a molecular navigation system that supercharges the ability of immune cells to “lock onto” cancer cells, targeting them for destruction. To do this, the T-cells are created with synthetic proteins designed to accurately seek and target cancer cells. These proteins were designed with the help of three different artificial intelligence tools.
In the body, T-cells kill cancer cells. However, they’re not always one hundred percent accurate in identifying the target cells. The researchers in Denmark genetically engineered T-cells to carry custom proteins on their surface. The proteins help to guide the T-cells to their target. In addition to recognizing cancer cells, the proteins can also recognize peptides from viral proteins and neoantigens, and direct T-cells accordingly.
The researchers used three AI tools to design their custom proteins. First, they fed the structure of the target cancer cell into RFdiffusion, an open source generative AI model used to generate protein structures. Previously, the researchers had trained the model with known protein structures and their amino acid sequences. Using this knowledge, RFdiffusion came up with protein shapes tailored to the shape of the target cells.
Researchers then used a second AI model to suggest strings of amino acids, which, when folded into 3D structures, would be likely to form the proposed protein shapes.
Finally, with the help of a third AI tool, the team sorted through tens of thousands of protein designs that could work, and narrowed the samples down to 44. These, they tested in the lab. Of these 44 designs, they chose one for their experiment. The results of the experiment showed that the engineered T-cells could locate and rapidly kill melanoma cells, and could also prevent the cancer from growing.
Researchers envision this technology being used to create diagnostic tools and custom therapies for a variety of different conditions, and potentially therapies tailored to individual patients. Current methods of doing this, which involve hunting through people’s cells to pick out T-cell receptors, are quite laborious, and can take a lot of time. According to Christopher Klebanoff, a medical oncologist and researcher at Memorial Sloan Kettering Cancer Center. The process can take months. And even then, one can end up with a very small number of candidate proteins, or even none at all.
According to medical biotechnologist Timothy Jenkins, the Danish team’s model allows for rapid creation of protein designs – as little time as a day or two, and a mere few weeks to test them in the laboratory. And that’s big news for the development of new therapies.
The feasibility of making protein “to order” goes back to Nobel-winner Baker’s original research in the early 2000s, when he used a program called Rosetta to comb through databases of known protein structures to look for features that could be useful in the construction of a new, hypothetical protein. Rosetta came up with a sequence of 93 amino acids that could be combined into a new, hypothetical protein called Top7. Baker’s team synthesized a gene coding for Top7 and spliced it into bacteria. The result showed that custom protein design was possible.
Today’s AI tools have the potential to make this process faster, more efficient, and more accurate than ever before.
It’s still a long road to developing specific therapies. However, according to Stanley Riddell, an immunotherapy researcher at Fred Hutch Cancer Center in Seattle, the team’s work demonstrates the power of artificial intelligence in designing usable and highly targeted synthetic proteins. Says Riddel, these AI models are “likely to generate a whole new class of therapeutics for a variety of diseases that will go beyond cancer.” Indeed, earlier this year, the Danish team reported similar advances in AI-assisted protein design that could lead to better antivenin treatments for snake bites.
Snake bites cause upward of 100,000 deaths per year. Presently, the only available treatment is an antivenin composed of polyclonal antibodies derived from the plasma of immunized animals. These are very expensive, and their efficacy can be limited. The Danish team’s work with AI-assisted protein design created designs that were stable, had high binding affinity, and effectively neutralised the three families of three-finger neurotoxins, protecting laboratory mice during a lethal neurotoxin challenge. The researchers foresee these stable, easily manufactured toxin-neutralizing proteins as one day providing the basis for safer, cheaper, and more widely available antivenins. Taking a wider view, they see the potential of the technology to democratize treatment for a variety of tropical diseases, particularly neglected ones.
AI design allows for unprecedented abilities to quickly generate huge numbers of designs and sort through them with lightning speed and precision. Many predict it will revolutionize not only therapy development, but synthetic biology, drug development, sustainable biotechnology, and more. Data driven modelling using generative AI, many believe, could revolutionize molecular science as a whole.
Although the results of the experiments are exciting, researchers are still a ways away from implementation in humans. First, there will need to be more laboratory tests, and then animal testing, which could take years before the technology moves on to human trials. But the potential for the quick, accurate development of inexpensive and potentially individualized therapies is immense.