December 1, 2020 18:11:01
Google’s artificial intelligence unit has taken a giant step to predict the structure of proteins, potentially decoding a problem that has been described as similar to genome mapping. DeepMind Technologies Ltd’s AlphaFold reached the “solution” threshold in the recent Critical Assessment of Structure Prediction. The event began in 1994 and is held every two years to speed up research on the topic.
The different folds in a protein determine how it will interact with other molecules, and understanding them has an impact on discovering how new diseases like Covid-19 attack our cells, designing enzymes to break down pollutants, and improving yields.
DeepMind became a subsidiary of Google after its acquisition in 2014 and is best known for its AI for gamers, learning to beat Atari video games and beat world-famous Go players such as Lee Sedol. The company’s ambition was to develop artificial intelligence that can be applied to wider problems, and so far it has created systems that make Google’s data centers more energy efficient, identify eye diseases from scans, and generate human speech.
“These algorithms are now becoming strong and powerful enough to be applied to science problems,” said Demis Hassabis, CEO of DeepMind, in an interview with reporters. After four years of development work, “we have a system that is accurate enough to actually be of biological significance and relevance to biologists.”
DeepMind is currently looking for ways to offer scientists access to the AlphaFold system in a “scalable way,” said Hassabis.
CASP researchers analyzed the shape of the amino acid sequences for a set of approximately 100 proteins. Competitors were given the sequences and entrusted with predicting their shape. The AlphaFold score was almost perfectly in line with the CASP analysis for two-thirds of the proteins, compared to around 10% from the other teams, and was better than what DeepMind achieved two years ago when it won a competition on first entry, accurately predicting structure 25 with 43 proteins.
Hassabis said his inspiration for AlphaFold came from “citizen science” attempts to find unknown protein structures such as Foldit, who presented the problem to amateur volunteers in the form of a puzzle. During the first two years, human players proved surprisingly good at solving puzzles and eventually discovered a structure that confused scientists and designed a new enzyme that was later confirmed in the laboratory.
“It often took years of experimentation to determine the structure of a single protein,” said Janet Thornton, Emeritus Director of the European Bioinformatics Institute and one of the pioneers in using computational methods to understand protein structure. ‘A better understanding of protein structures and the ability to predict them with a computer means a better understanding of life, evolution and, of course, human health and disease.’
? Indian Express is now on Telegram. Click here to join our feed (@indianexpress) and stay updated on the latest headlines
For all the latest tech news, download the Indian Express app.