DeepMind, the AI technology company that’s part of Google parent Alphabet, has achieved a significant breakthrough in AI-based protein structure prediction. The company announced today that its AlphaFold system has officially solved a protein folding grand challenge that has flummoxed the scientific community for 50 years. The advance in DeepMind’s AlphaFold capabilities could lead to a significant leap forward in areas like our understanding of disease, as well as future drug discovery and development.
The test that AlphaFold passed essentially shows that the AI can correctly figure out, to a very high degree of accuracy (accurate to within the width of an atom, in fact), the structure of proteins in just days — a very complex task that is crucial to figuring out how diseases can be best treated, as well as solving other big problems like working out how best to break down ecologically dangerous material like toxic waste. You may have heard of “Folding@Home,” the program that allows people to contribute their own home computing (and formerly, game console) processing power to protein folding experiments. That massive global crowdsourcing effort was necessary because using traditional methods, portion folding prediction takes years and is extremely expensive in terms of straight cost and computing resources.
DeepMind’s approach involves using an “Attention-based neural network system” (basically a neural network that can focus on specific inputs in order to increase efficiency). It’s able to continually refine its own predictive graph of possible protein folding outcomes based on their folding history, and provide highly accurate predictions as a result.
How proteins fold — or go from being a random string of amino acids when originally created, to a complex 3D structure in their final stable form — is key to understanding how diseases are transmitted, as well as how common conditions like allergies work. If you understand the folding process, you can potentially alter it, halting an infection’s progress mid-stride, or, conversely, correct mistakes in folding that can lead to neurodegenerative and cognitive disorders.
DeepMind’s technological leap could make accurately predicting these folds a much less time and resource-consuming process, which could dramatically change the pace at which our understanding of diseases and therapeutics progresses. This could come in handy to address major global threats, including future potential pandemics like the COVID-19 crisis we’re currently enduring, by predicting viral protein structures to a high degree of accuracy early in the appearance of any new future threats like SARS-CoV-2, thus speeding up the development of potential effective treatments and vaccines.