Harvard Medical School biologist Mohammed AlQuraishi has used the latest machine-learning technology to identify structural patterns in well-understood proteins and then apply them to other proteins.
The results are not precise enough for protein folding applications, such as finding new drugs, but they are at least one million times faster than traditional computer techniques. And this is just a first crack in a technology that can be improved and combined with other modeling techniques.
It is an example of how AI, despite fears of repercussions such as cutting police states or eliminating human jobs, has the potential to do so, including improving medicine.
"We now have a completely new perspective to explore protein folding," AlQuraishi said in a statement on Wednesday. "We've just started scratching the surface."
AI today mostly refers to neural network technology that is loosely based on human brains, and it revolutionized everything from voice commands and face recognition to software debugging and powering on windshield wipers. AI models learn patterns from real training data. This means that man does not have to give specific instructions on how to define what that sounds like when someone says, "Alexa, what's the weather like today?"
In every other form of life on earth, DNA strands contain instructions on how to assemble amino acids into long strands that become proteins. The laws of physics determine exactly how these threads collapse in close bundles, and the resulting surface structures are crucial for protein interactions in cells.
Precise modeling, as happens within a computer, quickly becomes difficult for larger proteins. That is, it's hard to understand what's going on with proteins. However, AlQuraishi believes that the AI technique may not only contribute to this understanding, but may also be used to construct new proteins that perform a particular task.
The results of AlQuraishi were published Wednesday in the journal Cell Systems.