By analyzing thousands of known proteins and their physical shapes, a neural network can learn to predict the shapes of others.[i]
Imagine you are a biologist who has spent your career devoted to a puzzle that could change the future of medicine—yet reaching your deathbed without the answer? Perhaps, as you breathe your last, you gain some consolation knowing your peers had no better results than you, and that younger generations will continue on after you are gone until success is finally achieved.
What if it turns out that instead of dying, during your lifetime a young whippersnapper called Artificial Intelligence suddenly pops up with the solution? As one dedicated to science, you would likely join in the elation shared by your fellow researchers the world over. But would you be left wondering how a computer managed to outrace all the human brains that came before?
In fact, the above scenario has occurred, as announced shortly after Thanksgiving, 2020. The puzzle was called “the protein folding problem,” a mystery first articulated around 1960 about the biology of protein molecules in the human body. The puzzle has three parts: (a) the folding code by which a sequence of amino acids that make up the protein takes a given shape; (b) what routes or pathways some proteins use to fold so quickly, and (c) the computational problem of how to predict a protein’s native structure from its amino acid sequence.[ii]
Brief background on protein
Protein molecules initially form as simple chains of amino acids, with specific chain sequences for each type of protein. These strings aren’t useful until they fold themselves into 3-dimensional shapes. Though this is a weak analogy, think of as a single skein of yarn that can’t keep you warm until knitting needles perform their magic to shape it into a sweater.
In the case of proteins, however, the body doesn’t need knitting needles. Instead, the protein strings naturally coil and fold themselves into unique 3-dimensional shapes in order to become useful biological actors. The unique structure into which a given protein folds “exposes a number of channels, receptors, and binding sites, and affects how it interacts with other proteins and molecules.”[iii] Its shape defines its function. Thus, folded proteins are the drivers that determine the very life behaviors of all organisms.
Artificial Intelligence Accurately Predicts Protein Structure
Over decades, the first two parts of the protein folding problem were gradually elucidated through sophisticated chemistry and physics laboratory analytics. That left the computational problem of predicting a protein’s structure based on its amino acid sequence. By 2008, there were even computer programs that could predict the structure of small molecules. But the puzzle facing computational biologists was how to crack the big nut for all proteins. The stakes were high:
This could help to (a) accelerate drug discovery by replacing slow, expensive structural biology experiments with faster, cheaper computer simulations, and (b) annotate protein function from genome sequences. … [P]rotein structure prediction has become as much a problem of inference and machine learning as it is of protein physics.[iv]
Well, all the brilliant computer scientists that have been grappling for nearly 50 years with inferences can indeed defer to machine learning. As reported in the New York Times, the predictability puzzle has been solved. “Now, an artificial intelligence lab in London has built a computer system that can do the job in a few hours — perhaps even a few minutes.”[v]
A computer lab, DeepMind, has developed a system called AlphaFold. DeepMind applied an Artificial Intelligence technology called a neural network to the problem of structure prediction. By modeling a mathematical system after the network of the brain’s neurons, and designing it specifically for protein folding, AlphaFold was “trained” using thousands of proteins. As a result, it needs only 2/3 of the time it would take previous computations to accurately predict the shape of a protein based on its amino acid sequence, and “its mistakes with these proteins are smaller than the width of an atom — an error rate that rivals physical experiments,” says the Times article.
AlphaFold has beat out other contenders in solving the protein folding problem, but its developers have not yet released it for prime time. Still, the research world is cheering its potential for rapidly developing or repurposing new drugs, and owes great gratitude to Artificial Intelligence for achieving a quantum leap in predicting the structure of folded proteins.
NOTE: This content is solely for purposes of information and does not substitute for diagnostic or medical advice. Talk to your doctor if you are experiencing pelvic pain, or have any other health concerns or questions of a personal medical nature.
[i] Metz, Cade. “London A.I. Lab Claims Breakthrough That Could Accelerate Drug Discovery.” New York Times, Nov. 30, 2020.
[ii] Dill KA, Ozkan SG, Shell MS, Weikl TR. The Protein Folding Problem. Annu Rev Biophys. 2008 Jun 9; 37: 289–316.
[iii] “Why is Protein Folding Important in Biology?” https://www.healthtechzone.com/topics/healthcare/articles/2019/12/03/443892-why-protein-folding-important-biology.htm
[iv] Dill, et al. Ibid.
[v] Metz, Ibid.