Protein folding has been a computer problem famous for decades: how do you discover the exact structure of these massive molecules that our DNA defines? Now, artificial intelligence is leading us to a much faster response.
Harvard Medical School biologist Mohammed AlQuraishi used the latest machine learning technology to detect well-understood protein structural patterns and then apply them to other proteins.
The results, while not accurate enough for protein folding applications such as the discovery of new drugs, are at least one million times faster than conventional computer techniques. And this is just a first flaw in a technology that can be improved and combined with other modeling techniques.
It is an illustration that AI, while fraught with fears about effects such as police complicity or the elimination of human jobs, has the potential to improve medicine, among other things.
"We now have a new vision from which we can explore the folding of proteins," AlQuraishi said in a statement on Wednesday. "We just started scratching the surface."
Today, AI most often refers to neural network technology based on human brains, and has revolutionized everything from voice commands and facial recognition to software debugging and the activation of windshield wipers. AI models learn patterns from real-world training data, an approach that means that humans do not need to do specific instructions, like trying to define what it looks like when someone says "Alexa, what's the weather like today?"
In humans and any other life on Earth, DNA strands contain instructions on how to assemble amino acids into long strands that transform into proteins. The laws of physics determine exactly how these chains collapse into tight bundles, with the resultant surface structures critical for the interactions of proteins within cells.
But modeling exactly how this will happen inside a computer makes it difficult to obtain larger proteins quickly. This means that it is difficult to understand what is happening with proteins. AlQuraishi, however, believes that the AI technique could not only help in this understanding, but also potentially be used to design new proteins that perform a specific job.
AlQuraishi's results were published Wednesday in the journal Cell Systems.