DeepMind's latest artificial intelligence (AI) software AlphaFold has won the Critical Assessment of Structure Prediction (CASP) competition by accurately predicting the 3-D structures into which proteins can be folded.
AlphaFold was adjudged the winner among a total 98 algorithms. It predicted the shapes of 25 out of 43 proteins; the first runner-up algorithm, in comparison, could predict only three of the 43 protein structures, according to The Guardian.
The contest was organised by the Protein Structure Prediction Centre, an institution supported by the U.S. National Institute of General Medical Sciences.
"For us, this is a really key moment," said Demis Hassabis, co-founder and CEO of DeepMind.
"This is a lighthouse project, our first major investment in terms of people and resources into a fundamental, very important, real-world scientific problem."
DeepMind is the AI firm owned by Google parent company Alphabet Inc.; DeepMind's AI algorithm is known for having attained the superhuman performance in chess and Chinese strategy game, GO.
In 2016, the algorithm proudly beat the world's top professional player Lee Sedol in the game. Later, DeepMind expressed its desire to apply its AI technology to address fundamental problems in science. The firm says it is exploring the application of AI technology in drug development domain.
Proteins are the fundamental molecules of life, and therefore it is vital for scientists to understand how proteins fold up. Human body can create a large number of proteins, ranging from tens of thousands to millions.
Proteins are made up of amino acids (20 different types), and can assume staggering number of shapes by twisting and bending between amino acids. Using AI to predict protein shapes can enable scientists to solve several problems impacting environment, health, and those involving biological systems.
According to DeepMind, AlphaFold can generate 3-D models of proteins with far more accuracy than any other algorithm. The algorithm was designed and trained to model target shapes from scratch, without using previously solved proteins as templates. The team used two different neural networks to predict the structures of proteins.
"We've not solved the protein folding problem, this is just a first step," Hassabis said.
"It's a hugely challenging problem, but we have a good system and we have a tonne of ideas we haven't implemented yet."
Ceres, located in the asteroid belt, has a carbonaceous-rich upper crust, SwRI study claims
The spacecraft found traces of hydrogen and oxygen molecules, known as hydroxyls, embedded in the rocky surface of the asteroid
The skeleton was unearthed more than 20 years ago in South Africa
Moon's dark side is mountainous, rugged and never visible from the Earth