Then strive for some breakthroughs and bite the bullet, Fudan professor Ma Jianpeng: On the cutting-edge track that our country cannot afford to lose
"Nature" magazine recently published a joint study by the Google Deepmind team and the AI drug research and development team, demonstrating a new protein prediction model-the third generation "Alpha folding". At the 79th Cultural Forum of Jiefang Daily held recently, when asked "When will such a heavyweight innovation appear in Shanghai?", Ma Jianpeng, dean of the Institute of Complex Systems and Multi-Scale at Fudan University, replied:
This is a very good question, and it’s something I often ask myself, and it’s something we’ve been doing.
Alphafold is a landmark event in the history of artificial intelligence. As we all know, when the computer defeated the world chess champion, everyone thought that artificial intelligence could not defeat humans in the field of Go. Later, AlphaGo defeated the top human Go player, but everyone still had some doubts, so Google chose a scientific problem, which is called "one of the most difficult scientific problems under the sun" - protein structure prediction, as a research direction.
In 2018, Deepmind announced for the first time that it had developed a tool, Alphafold, to accurately predict the folding structure of proteins. In 2020, Alphafold2 was upgraded and launched, which shocked the world.
The recent Alphafold3 has achieved a huge breakthrough: originally Alphafold2 could only predict pure proteins, but Alphafold3's prediction objects go beyond proteins and cover nucleic acids, small molecules, metal ions, etc. This is a huge improvement.
Director of the Institute of Multiscale Complex Systems at Fudan University.
So, what has our country done? In a particularly cutting-edge field like this, we must first firm our direction, stick to it without wavering, and stick to it. Secondly, if you go head-on, you will definitely not be able to defeat others, so you have to overtake in corners and strive for partial breakthroughs.
We know that the three-dimensional structure of a protein is composed of a main chain and a side chain. Alphafold2's main chain prediction is generally good, but the quality of the side chain prediction is not good enough, at least there is still a big gap between the accuracy required for drug design. We put a lot of effort into figuring out where the weaknesses of Alphafold2 are. In October last year, the Multiscale Institute of Complex Systems at Fudan University developed an algorithm called OPUS-Rota5, which can greatly improve the accuracy of protein side chain structure testing. When Alphafold3 came out, we had overall reached the level of Alphafold2.5. Even with Alphafold3, our side-chain structure testing accuracy still maintains the world's leading level.
It must be pointed out here that the practical use of a protein structure with an accurate main chain structure but large errors in the side chain structure is very limited, because almost all protein-related interactions cannot circumvent interactions with side chains. In short, the competition in the direction of Alphafold technology, which has been the biggest international breakthrough in the field of computational biology in recent years, is a track that our country cannot afford to lose because it involves the development of almost all fields of molecular biology and materials science.
Of course, colleagues in the AI field hope that one day we can make a breakthrough in the underlying logic, but this requires patience. You can't push scientists every day, "Have you done it?" because these topics are really, really difficult. Once a breakthrough occurs, it can subvert the whole world, and this world is not so easy to be subverted. However, given time, disruptive breakthroughs can be achieved, and there is no reason why breakthroughs cannot be achieved in Shanghai.
Another important point: Almost all of the recent huge breakthroughs in AI have been made by technology companies. Why? This is related to the research and development characteristics of AI. In colleges and universities, professors can maximize personal creativity, but some jobs, especially today's AI, require large groups to fight and cannot be completed by one or a few people. In a company, under the leadership of a leader, mobilizing various resources, including various talents, it is more likely to achieve breakthroughs.
