Recently, there was a saying circulating on the Internet about the "real reason" of structural biologist Yan Ning's return to China, saying that the advent of artificial intelligence systems such as AlphaFold 2 (AlphaFold 2) has made structural biologists face the dilemma of "unemployment". Someone wrote on Zhihu: "Professor Yan Ning saw AlphaFold like a brave and warlike tribal leader saw an aircraft carrier. It's not that Yan Ning can't do it, but the British DeepMind is too powerful. ”
Is this the case? The reporter of Jiefang Daily Shangguan News interviewed Professor Ma Jianpeng, a computational biologist and dean of the Multiscale Institute of Complex Systems of Fudan University. He led the team that has developed the OPUSFold system that functions similarly to AlphaFold 2. He bluntly said: "AI (artificial intelligence) has put first-class structural biologists out of work, which is the most ridiculous claim I have ever heard." ”
AI helps pick the "crown jewel"
"Alpha Fold 2" is a product of Google's DeepMind, similar to "Alpha Go", which is an artificial intelligence system using machine learning technology. At the International Protein Structure Prediction Competition held in 2020, "Alpha Fold 2" won the championship, and its predicted three-dimensional structure of protein was only slightly different from the experimentally determined structure, and was named one of the "Top Ten Scientific Breakthroughs of 2020" by Science magazine.
The protein structures of the "alpha fold 2" prediction (blue) and experimental assays (green) match almost perfectly. Source: DeepMind
Why use artificial intelligence systems to predict the three-dimensional structure of proteins? Ma Jianpeng explained that proteins are made by folding a series of amino acids. The amino acids linearly arrange into a long chain, which is placed in water, and the entire chain folds into a stable three-dimensional structure in microseconds to milliseconds. The study of how long chains of amino acids spontaneously fold into three-dimensional structures, referred to as "protein folding", is regarded as the "crown jewel" of modern molecular biology because of its importance and complexity. In the field of application, the basis of small molecule drug development is protein structure analysis, and only by exploring the "three-dimensional map" of the target protein can we find the target of the drug acting on the protein.
Determining amino acid sequences is relatively easy for scientists, but resolving protein structures is difficult because protein structures depend on the interaction forces between the atoms of several thousand amino acids. Based on known amino acid sequences, the amount of computation required by computers to predict protein structure is difficult for even the world's fastest supercomputers to handle.
With the rise of artificial intelligence technologies such as deep learning and reinforcement learning, computational biology has developed by leaps and bounds. Systems such as "Alpha Fold 2" have the ability to accurately predict structures based on amino acid sequences after learning a large number of protein structures determined experimentally. This year, Deep Thinking released a dataset update saying that Alpha Fold 2 has predicted almost all known proteins.
Academician Yan Ning responded to the rumors through Weibo
Now that AI systems can accurately predict protein structure, will structural biologists face the dilemma of "job loss"?
According to reports, structural biology is an interdisciplinary discipline that studies the three-dimensional spatial structure, dynamic process and biological function of biological macromolecules. Analyzing the three-dimensional structure of various proteins is a major business of structural biologists. As an internationally renowned structural biologist, Professor Yan Ning has worked at Tsinghua University and Princeton University, is a foreign member of the National Academy of Sciences and an academician of the American Academy of Arts and Sciences. This month, she revealed that she had submitted her resignation to Princeton University and would soon return to China full-time to help found the Shenzhen Academy of Medical Sciences.
Professor Yan Ning gave a speech. Source: Visual China
For the online rumor that Yan Ning's "real reason" for returning to China, she has responded through Weibo: In the field of voltage-gated sodium ion and calcium ion channels she studied, "Alpha Folding 2" learned multiple biological structures that she led her team to analyze, and last year's prediction accuracy reached the level of Yan Ning's team in 2017, but this year there was no progress. "The AI team made predictions, we did experiments to test the interaction of new small molecules with proteins, and none of the predictions so far were correct."
"Combination of wet and dry" has become a biological trend
Ma Jianpeng said that "Alpha Fold 2" is far from the ability to replace structural biologists. At present, it can only predict the structure of single-chain proteins, and basically does not have the function of predicting the structure of multi-chain proteins. Moreover, in terms of single-chain protein prediction, because artificial intelligence prediction is based on comparative learning of known protein structures, it is relatively accurate to predict the protein structure of its homologous, but in the face of proteins with "orphan sequences" (unique amino acid sequences), "alpha fold 2" is often unable to accurately predict.
In addition, in terms of protein side chain prediction, "Alpha Fold 2" also has a large room for improvement. In 2021, the Institute of Multi-scale Complex Systems of Fudan University published a paper in the British "Bioinformatics Bulletin" reporting that the "work folding" they developed was higher than the "alpha folding 2" in terms of protein side chain prediction accuracy. According to reports, the three-dimensional structure of the protein is built by the main chain and the side chain. Most of the binding of drug molecules and proteins is achieved by interacting with amino acid side chains, so the accurate prediction of side chain structure by artificial intelligence systems is of great value to new drug research and development.
Blue is the natural conformation of the protein side chain, and red is the "work folding" prediction.
It can be seen that artificial intelligence will not make structural biologists "unemployed", and the two are not replacing relationships, but complementing each other. "AlphaFold 2 is beneficial to a first-class experimental structural biologist like Yan Ning." Ma Jianpeng said, "Experimental structural biologists also want to use computer modeling, and software such as AlphaFold 2 and OPUSFold can speed up modeling and improve the efficiency of protein structure analysis." ”
Today, "dry-wet combination" has become a trend in structural biology research. For a long time, the "dry laboratory" that carried out computational biology research was a supporting role in biology. With the rise of artificial intelligence, this supporting role has gradually grown into a protagonist, more closely integrated with the "wet laboratory" where experimental biologists work, to explore the mysteries of the molecular structure of life.
"True researchers are willing to embrace technological progress and are good at using various technologies to explore and answer questions that interest them." Yan Ning said, "I expect AI to become more and more powerful. ”
Editor-in-chief: Huang Haihua Source: Visual China Photo editor: Zhu Jun
Source: Author: Yu Taoran