[ad_1]
A long-standing and incredibly complex scientific problem related to the structure and behavior of proteins has been effectively solved by a new artificial intelligence (AI) system, the scientists report.
DeepMind, the UK-based artificial intelligence company, has wowed us for years with its parade of ever-advancing neural networks continually beating humans in complex games like chess and Go.
However, all those incremental advances were about much more than mastering recreational amusements.
In the background, the DeepMind researchers sought to persuade their AIs to solve much more important scientific puzzles, such as finding new ways to fight disease by predicting infinitesimal but vitally important aspects of human biology.
Now, with the latest version of their AlphaFold artificial intelligence engine, they seem to have achieved this very ambitious goal, or at least brought us closer than scientists have.
For about 50 years, researchers have struggled to predict how proteins achieve their three-dimensional structure, and it is not an easy problem to solve.
The astronomical number of potential configurations is so staggeringly huge, in fact, that the researchers postulated that it would take longer than the age of the Universe to sample all possible molecular arrangements.
However, if we can solve this puzzle, known as the protein folding problem, it would be a major advance in scientific capabilities, greatly accelerating research efforts in things like drug discovery and disease modeling, and it would also lead to new applications far beyond health.
For that reason, despite the scale of the challenge, researchers have been collaborating for decades to advance in developing solutions to the problem of protein folding.
A rigorous experiment called CASP (Critical Evaluation of Protein Structure Prediction) began in the 1990s, challenging scientists to devise systems capable of predicting the esoteric puzzles of protein folding.
Now in its third decade, the CASP experiment appears to have produced its most promising solution yet: DeepMind’s AlphaFold offers predictions of 3D protein structures with unprecedented precision.
“We’ve been stuck with this problem, how proteins fold, for almost 50 years,” says CASP co-founder John Moult of the University of Maryland.
“Seeing DeepMind produce a solution for this, after having personally worked on this problem for so long and after so many stops and starting to wonder if we would ever get there, is a very special moment.”
In the experiment, DeepMind used a new deep learning architecture for AlphaFold that was able to interpret and calculate the ‘spatial graph’ of 3D proteins, predicting the molecular structure that underpins their folded configuration.
The system, which was trained by analyzing a database of approximately 170,000 protein structures, contributed its unique abilities to this year’s CASP challenge, called CASP14, achieving an average score on its predictions of 92.4 GDT (Test of global distance).
That’s above the ~ 90 GDT threshold that is generally considered competitive with the same results obtained through experimental methods, and DeepMind says its predictions are only off by about 1.6 angstroms on average (about the width of an atom).
“I almost fell out of my chair when I saw these results,” says genomics researcher Ewan Birney of the European Laboratory for Molecular Biology.
“I know how rigorous CASP is – it basically guarantees that computational modeling should work on the challenging task of From the beginning protein folding. It was humbling to see that these models could do that with such precision. There will be many aspects to understand, but this is a great advance for science. “
It’s worth noting that the research has yet to be peer-reviewed or published in a scientific journal (although DeepMind researchers say it is on its way).
Still, experts familiar with the field are already acknowledging and applauding the progress, even if the full report and detailed results have yet to be seen.
“This computational work represents a staggering advance on the problem of protein folding, a great 50-year challenge in biology,” says structural biologist Venki Ramakrishnan, president of the Royal Society.
“It has happened decades before many people in the field would have predicted it.”
The full findings are not yet published, but you can view the research summary, “High-precision protein structure prediction using deep learning,” here, and find more information about CASP14 here.
