In biology, structure dictates function.
That’s certainly the case with protein-encoding genes: the sequence of bases in a specific gene dictates the sequence of bases in an intermediate molecule called messenger RNA, which in turn, dictates the selection and sequence of amino acids in the protein and the form and function of that protein.
Most proteins are complex chains of hundreds of amino acids. By now many of us are familiar with computer-generated 3-D models of the coronavirus’s spike protein, accompanied by arrows pointing to this or that mutation or combination of mutations, which make a variant more transmissible, potentially more lethal, or perhaps, in the case of the Omicron variant, less likely to cause moderate to severe symptoms.
Recent studies illustrate the importance of identifying potentially troublesome mutations and understanding how they affect the shape and charge of the spike protein, and so alter the virus’s ability to infect host cells and evade the host’s immune responses.
This type of biological detective work has become a universe where physics, chemistry and biology fuse together. That fusion was on display for recent Nobel Prizes in physics and chemistry.
For example, Nobel Prizes in chemistry in 2018 and 2021 highlighted methods for developing more effective organic, that is protein, enzymes, bioengineering potent antibodies (also proteins) for treating cancer and discovering the active sites in enzymatic proteins most responsible for those enzymatic properties.
And the 2021 prize in medicine-physiology was awarded for truly elegant studies that illustrated how the shape of receptor proteins translates mechanical stimuli involving the skin, muscle and deeper tissues into electrical signals in the peripheral nervous system.
But until recently, figuring out how the type and order of amino acids in proteins affect the shape and therefore function of proteins was a laborious, expensive, and time-consuming task. The search began in the 1950s using X-ray crystallography, to which was later added nuclear MR spectroscopy, and most recently and topping the list for expense, cryo-electron microscopy.
As early as the 1960s scientists also began to harness computer algorithms to help them figure out how the sequence of amino acids dictates the shape of proteins.
To speed up the process, scientists launched a biennial competition to pick the best algorithms. To no one’s surprise, the algorithms helped, but not enough – a major advance was needed.
In 2018, Google’s sister company DeepMind entered the game with self-learning programs, the first of which was called AlphaFold and, in updated form, AlphaFold2.
The latter and similar software developed by others such as RoseTTA-Fold were trained on a database of hundreds of proteins whose 3-D structures had been laboriously worked out beforehand using conventional techniques.
Once trained on the database, AI programs soon learned to work out the 3-D structures of hundreds of proteins with a precision matching standard techniques. The difference was that AI took far less time to solve the origamic challenge of relating amino acid sequences to 3-D shapes for proteins.
For medicine, the rewards are huge because the sequences of amino acids dictate the function of thousands of proteins from receptors to antibodies, to enzymes and a host of other proteins which play key roles in the brain, muscles, immune system, and other systems.
One of the major payoffs may come with the design of novel drugs tailored to prevent the accumulation and/or clear the brain of the structurally and hence functionally distinctive misfolded protein aggregates found in a wide spectrum of neurodegenerative diseases.
The latter include Alzheimer’s disease, Pick’s disease, traumatic encephalopathy, cortical basal degeneration, progressive supranuclear palsy as well as infectious forms of neurodegenerative diseases such as bovine-encephalopathy and Creutzfeldt-Jacob disease, and other as-yet untreatable encephalopathies.
That’s one of several reasons why the introduction of AI as a powerful tool for solving the riddles of protein structure by John Jumper and his team was chosen by the journal “Nature” as one of its top 10 advances in science in 2021.
It was yet another triumph for artificial intelligence with many more expected in science and medicine in coming years and potentially a big win for the recognition and early treatment of the dementias and many other neurodegenerative diseases, which so far, have stubbornly stymied physicians and scientists.
Dr. William Brown is a professor of neurology at McMaster University and co-founder of the InfoHealth series at the Niagara-on-the-Lake Public Library.