Computer scientists at DeepMind, a lab owned by Alphabet, Google’s parent company, have developed an AI that has made a monumental leap in solving “the protein folding problem.” This conundrum has plagued biologists for decades, and solving it can accelerate countless other studies in medicine, agriculture, and conservation, to name a few.
Proteins are crucial “macromolecules” built from strings of molecules called amino acids. These strings of molecules are built inside of each cell by reading an organism’s DNA and blocking together long chains of amino acids. Of the 20 amino acids that make up proteins, each one chosen depends on the DNA sequence being read, and each amino acid has special properties that cause the long chain of molecules to fold over itself into a very specific, highly complicated shape, which gives the protein its function.
Understanding the shape and how it relates to the function of proteins is so important because they are responsible for most biological functions in the body, including immune response, digestion, energy production, muscle contraction, blood clotting, and countless more.
The interest of many research groups revolves around synthetic proteins used in medicines. Knowing the order of amino acids in a protein is easy enough for researchers if they know the DNA sequence that is used as its template, but knowing its shape can dramatically expedite the process of determining how it will function in people, bacteria or viruses. Among other mysteries this can unravel, think of vaccine research, a topic on everyone’s mind presently.
DeepMind’s neural network can quickly and reliably determine a protein’s shape, given the string of amino acids that comprise the protein.
Scientists have been working hard to crack this nut for over 50 years. A competition called the Critical Assessment of Structure Prediction (CASP) was formed 25 years ago to bring together this research, compare results and spur on investigation.
No one has really come close to solving this problem until DeepMind came onto the scene. Tested on a variety of proteins, DeepMind’s neural network was able to accurately assume the folded structure of a protein with the precision of lab-based experiments that typically take months or years to solve.
It should be noted that experts say solving the protein folding problem speeds up just the initial stage of the long process of safely developing a medication.
This has also apparently emerged too late to affect vaccine research into the coronavirus but could be used down the line to stem future pandemics, determine whether existing medications can be used on novel infections, and research genetic diseases such as Alzheimer’s.
John Moult, a University of Maryland professor that co-created CASP back in 1994, said about this new development, “I always hoped I would live to see this day.”