Revolution Breakthrough: AI May Accurately Predict Protein Structure Based on Amino Acid Sequence

The artificial intelligence (AI) network developed by Google AI makes DeepMind stand out and has made a huge leap in solving one of the toughest challenges in biology-determining the 3D shape of a protein from its amino acid sequence. DeepMind's program called AlphaFold outperformed 100 other teams in the biennial protein structure prediction challenge called CASP, which is short for Key Evaluation of Structure Prediction.

In some cases, AlphaFold's structure predictions are no different from those determined using "gold standard" experimental methods (such as X-ray crystallography and cryo-electron microscopy in recent years). AlphaFold is unlikely to close a laboratory like Shi Yigong that uses experimental methods to solve protein structures. Lupas said: "This will enable a new generation of molecular biologists to ask more advanced questions. This will require more thinking and fewer pipetting operations.”

The ability to accurately predict protein structure from amino acid sequence will bring huge benefits to life sciences and medicine. This will greatly speed up the work of understanding cellular components and make faster and more advanced drug discovery possible. AlphaFold was among the best in the last (2018) CASP, which was the first year that London-based DeepMind participated. However, this year, the institution's deep learning network is ahead of other teams, and scientists say that their performance is incredible and can herald a biological revolution.

"This is a game-changer," said Andrei Lupas, an evolutionary biologist at the Max Planck Institute for Developmental Biology in Tubingen, Germany, who evaluated the performance of different teams in CASP. AlphaFold has helped him find the protein structure that has plagued his laboratory for ten years, and he hopes that this structure will change the way he works and solve the problems. In some cases, AlphaFold's structure predictions are no different from those determined using "gold standard" experimental methods (such as X-ray crystallography and cryo-electron microscopy in recent years). Scientists say that AlphaFold may not eliminate the need for these laborious and expensive methods, but AI will make it possible to study biology in new ways.

"This is a problem that I cannot solve in my entire life," said Anet Thornton, a structural biologist at the European Molecular Biology Laboratory-European Institute of Bioinformatics in Hinkton, UK, and a former CASP assessor. She hopes that this method can help elucidate the functions of thousands of undissolved proteins in the human genome and clarify the different disease-causing gene variants between people.

The performance of AlphaFold also marks a turning point for DeepMind. The company is known for using AI to master Go and other games, but its long-term goal is to develop programs that can achieve a wide range of human intelligence. Hassabis said that coping with huge scientific challenges, such as protein structure prediction, is one of the most important applications that AI can achieve. "In terms of real-world impact, I do think this is the most important thing we do."