Half of this year's Nobel Prize in Chemistry goes to AI: the Nobel Prize has entered the era of AI, and focus and purity will never go out of style

AI has won this year! Following two artificial intelligence (AI) pioneers winning this year's Nobel Prize in Physics, the AI-powered protein structure prediction model "AlphaFold2" won one-quarter of the 2024 Nobel Prize in Chemistry tonight. Some commentators believe that a revolutionary paradigm shift in scientific research has been "stamped and confirmed" by the Nobel Prize Committee, and the Nobel Prize is about to enter the AI era.

According to the Sweden Royal Academy of Sciences, half of the 2024 Nobel Prize in Chemistry will be awarded to David · Baker, a professor at the University of Washington, "for his contributions to computational protein design"; The other half jointly awarded AI scientists Demis · Hasabis and John · Jiangper of Google's DeepMind "for their achievements in protein structure prediction."

20 years of "long-distance running" and revolutionary "kick in the door"

"AlphaFold" won the Nobel Prize, and the industry seems to have a premonition, which can be described as expected. However, at the moment when the Nobel Prize in Chemistry was announced tonight, people still wondered: why was the achievement "computational protein design" at the top of the list and winning 1/2 of the awards?

"In the field of protein design, David · Baker is a Taishan Beidou-level existence, and he has continued to carry out fundamental development for more than 20 years, so that human understanding and control of protein have reached an unprecedented height." Xu Wenqing, a professor at ShanghaiTech University, said David has developed a series of computer methods to create many proteins with completely new functions that did not exist before, which was considered "impossible" in the past.

Before AI intervened, designing proteins from scratch was an extremely painstaking task with a low success rate. However, in the past 20 years, structural biology has accumulated a huge amount of data, which has paved the way for AI-based protein structure prediction and protein design.

In 2020, "AlphaFold2" kicked off a revolutionary "kick in the door". In the view of Professor Shan Bing of the iHuman Research Institute of ShanghaiTech University, this can be said to be the product of the convergence of "time, place, and people". Through the ingenious deep learning of a large number of protein structures and sequences accumulated over many years, this AI model has improved the prediction accuracy of protein structures from less than 40% to more than 90% in just one or two years, and can predict almost all known protein structures.

In fact, predicting the three-dimensional structure of proteins based on amino acid sequences has long been considered "one of the most difficult scientific problems under the sun". DeepMind has published the code for "AlphaFold2", which has been used by more than 2 million people from 190 countries to date. What used to take years to obtain protein structure now takes only a few minutes.

Jiayi Dou, a Ph.D. student at David's and a researcher at the School of Life Science and Technology at ShanghaiTech University, mentioned that "AlphaFold2" has become one of the commonly used scientific research tools in protein design work, and David's lab is also using it. Xu Wenqing said that using this model, the success rate of protein design has made a qualitative leap - now the protein structure is designed from scratch, and each project can always pick out a few better designs for further optimization for the desired part of the function.

AI won the Nobel Prize, and this year is just the beginning

Since the advent of "AlphaFold", Deepmind's model has been regarded as a model for "AI for Science".

"Protein structure prediction can be said to be a 'holy grail' problem in molecular biology." Zhang Qiangfeng, an associate professor at Tsinghua University's School of Life Sciences, believes that "AlphaFold2" uses AI end-to-end neural network algorithms to solve this "problem that has plagued the scientific community for 50 years" to a certain extent. Last year, the developer of "AlphaFold" won the Lasker Award, which means that it has been universally recognized by the scientific community.

Professor Ma Jianpeng, Dean of the Institute of Multiscale Research for Complex Systems at Fudan University, said: "I once proposed that AlphaFold was a Nobel Prize-level contribution, but I didn't expect it to win so quickly. In his opinion, from the debut of "AlphaFold" in 2018, to the "AlphaFold2" being named one of the top ten scientific breakthroughs in 2020 by Science magazine in United States, and then to the "AlphaFold3" released in May this year, every version update can be said to be a Nobel Prize-level leap. In particular, "AlphaFold3" directly changed the core architecture of the previous generation, replacing the very important "structural module" of the previous generation with "diffusion module".

In Xu Wenqing's view, even with the development of "AlphaFold3", the interaction between proteins and some small molecules, nucleic acids, and even protein modification functions are added, AI still has a lot of room for development in antibody and drug design, small molecule dynamic simulation, and in vivo complex structure simulation.

Zhang Qiangfeng believes that this just reflects the footprint of AI accelerating into the mainstream scientific community - scientists are putting forward a steady stream of demand for AI, "I believe that AI will become the core tool of scientific exploration in the future, and will also get more Nobel Prizes, this year is just the beginning."

Obsessed with important research and big questions

Whether it is John · Jiangpo after 85, or David · Baker, who is over the age of sixties, in the eyes of those who are familiar with them, they all have the characteristics of focus and purity. In the frontiers of science, this seems to be a quality that never goes out of style.

For the past 20 years, David has published several papers in Nature, Science, and Cell every year. Xu Wenqing, who had worked with David at the University of Washington for many years, often walked by his office and often saw David thinking about a problem, "always looking focused on the most important research."

In Dou Jiayi's mind, David has always maintained the academic youth of a "senior graduate student", which may be the secret of his mentor's perennial "high production" - he not only personally leads projects and experiments, but also asks students for advice on topics, "He rarely goes out for meetings, can see him in the laboratory almost every day, and likes to go to the mountains around Seattle to climb when he has time, which can be called the camping 'living map' there."

Prior to his bachelor's degree and Ph.D., John · worked at the Deshaw Institute in New York for three or four years, working in a project team led by a single soldier. "He's quiet, gentle but thoughtful." Shan Bing felt that Jiang Po was the kind of scientist who "likes to do big problems and solve all problems in one package", and "I have to say that Hassabis is quite bold and dares to reuse young people who seem to have little qualifications, which makes John stand out in just a few years."

Of course, DeepMind's success is inseparable from Google's well-funded "big team operation". Ma Jianpeng believes that for a basic research from 0 to 1, it is necessary to give sufficient financial support and development space. He said that the potential in the field of protein prediction is endless, and it is necessary to increase investment and work for a long time.