DeepMind tackles 50-year-old challenge with AI software to boost drug discovery: 5 details

The artificial intelligence system developed by Alphabet's DeepMind solved a 50-year-old challenge to help accelerate disease treatment and drug discovery, according to a Nov. 30 company blog post.

Five details:

1. The "protein folding problem" has been a complex challenge in biology for a half-century that Alphabet's AlphaFold solves by organizing the biennial critical assessment of protein structure prediction. Scientists have tried to solve the problem for years using techniques like nuclear magnetic resonance and X-ray crystallography, which are time intensive and require expensive equipment.

2. AlphaFold can predict the shape proteins will fold into with more accuracy than other technology, according to the blog post. The system achieved a median score of 92.4 Global Distance Test, which measures the percentage of amino acid residues within a threshold distance from the correct position.

3. The team behind AlphaFold developed an attention-based neural network system that predicts the underlying physical structure of the proteins and can determine the structures within "a matter of days," according to the post.

4. Researchers tested the system on about 170,000 protein structures from the publicly available data along with a large database that contained protein sequences of unknown structures. The system ran over a few weeks to produce results.

5. AlphaFold's development team is optimistic that the technology can help identify malfunctioning proteins for more precise drug development and treatment. It could also be used as a tool in future pandemics; the technology predicted multiple protein structures of SARS-CoV-2 virus that were previously unknown.

"These exciting results open up the potential for biologists to use computational structure prediction as a core tool in scientific research," said the AlphaFold team in the post. "Our methods may prove especially helpful for important classes of proteins, such as membrane proteins, that are very difficult to crystallize and therefore challenging to experimentally determine."

More articles on health IT:
9 recent big tech partnerships in healthcare: Amazon, Google, Microsoft
Johns Hopkins' Dwight Raum on the CIO's dual role with COVID-19 vaccines
California medical center server issue exposed patient data for 4.5 years