In classic Asimov fashion, the ending of the short story “The Last Question” offers an unexpected twist to the tale of how the Universe came to be formed. An important player in this story is AC, or Artificial Computer. This faithful servant of mankind persists through the trillions of years that human civilization exists; first on Earth, then on other planets, until finally occupying the entirety of the Universe. Until one day, humanity exists no more, and the servant becomes the Creator.
Artificial Intelligence taking control of humans has long been a favourite theme among authors and film-makers. The depictions have often been gory (read the Terminator series) and paint a bleak picture of human-AI interaction. In this article, we will be exploring these themes and breaking down some concepts of AI with the help of an example that has been creating ripples in the world of biological research: AlphaFold. For the uninitiated, AlphaFold is an artificial intelligence software which was created by Google’s DeepMind, an artificial intelligence research laboratory headquartered in King’s Cross, London. In a nutshell, this software was developed with the goal of cracking the problem of protein folding. The ‘magic’ happened here in 2022, when AlphaFold 2 was able to successfully predict the structures of over 200 million catalogued proteins. But what does this mean for science?
The process of protein folding can be broken down to a few simple steps: First, the linear chain of amino acids produced must be folded appropriately into a functional 3-dimensional structure. Like a cargo product requiring the right packaging, the protein too undergoes “packaging” or modifications. Finally, the prepared cargo product needs to reach its destination, and it does so with an address. In this case, the address is a specific signalling sequence that is attached to the modified protein, and it is successfully targeted to its designated location in the cell. What’s the catch in this full-proof method though? On the outset, this well-orchestrated process seems nearly perfect…except for one process: protein folding. The transition from a linear chain of amino acids to a complex, three-dimensional structure is far from simple, and here is the reason why: An unfolded polypeptide structure has a large number of possible conformations due to the bonds present in the linear chain, and yet only one conformation is its true, functional form. Just like Dr. Strange, who examines all the million outcomes of the Avengers’ war against Thanos and predicts a single, successful outcome for them, an unfolded polypeptide would need to do the same. Not surprisingly, this is a challenging task prone to error.
A protein going through millions of possible conformations before finalising on ‘the one’ would theoretically take very long. Yet in reality, proteins perform this function with a snap of their fingers! The final conformation attained is the least energetic one thermodynamically. And yet, as we age, this fundamental process seems to go awry and leads to degenerative diseases such as Alzheimer’s, Parkinson’s and even cancer. The reason for this remains unknown. What can be known, is the protein structure, which is deciphered using techniques such as X-ray diffraction, Nuclear Magnetic Resonance (NMR), cryo-Electron Microscopy etc. A better understanding of this structure would lead to a better understanding of targeting it using various drugs. Yet, why do we not have a ‘magic drug’ in market for these diseases, that still remain elusive to cure? The reason is the time it takes for research to land the precise structure with great accuracy. From performing multiple experiments to piecing together the structure and publishing the findings, is a long journey.
That is, until the advent of AI. In 2018, AlphaFold stood first in CASP (Critical Assessment of Techniques for Protein Structure Prediction), a biennial challenge that is organized for research groups to compete and test the accuracy of their predicted data of the protein against real experimental results. In 2020, AlphaFold 2 was introduced which showed exceedingly precise accuracy. AlphaFold 3 has come up in 2024, which is not only able to predict protein structures, but also interactions with other biomolecules, such as DNA and RNA, in addition to predicting the protein’s capability to bind to certain drugs. So not only has AI solved a 50 year old scientific problem, it has also given us a potential cure for some of the most debilitating diseases. Or has it?
Often, a new protein’s discovery requires it to be classified under a protein family. For proteins with lesser relatives, it becomes hard for AlphaFold to predict its structure. This is because the neural network behind AlphaFold makes its predictions by drawing comparisons from the protein’s closest relatives. Thus, if a new protein that plays a role in a disease is discovered, it would still take a while for the AI to come up with an accurate prediction. But we sure are getting there! AI has been referenced in pop culture for a long time, with Vision and Ultron of Marvel fame playing the good cop-bad cop roles. Only time will tell if we can come up with our own real-life, life-saving bots that would not only predict the protein structure, but also predict its variants in disease conditions and help in early diagnosis of at-risk patients.
Writer and Editor
Luminaa Anandh
Illustrator
Jayakrishnan Nair