On the evening of October 9, Beijing time, the paper of Professor Ma Jianpeng's team of Fudan University was published in the internationally renowned scientific journal Nature - Method (impact factor: 47.99), which introduced the structural biology algorithm OPUS-DSD developed by them, which can analyze the structure of biological macromolecules such as proteins and nucleic acids that are defective due to the inability of traditional methods in cryo-electron microscopy, and can also efficiently and accurately distinguish the conformational distribution of flexible domains in biological samples. The advent of this new method will help scientists establish high-precision biological macromolecular structure models and solve the problem of new drug development failures caused by inaccurate structure of target proteins.
Michael Levitt, Nobel laureate in chemistry and honorary dean of the Institute for Multi-scale Complex Systems at Fudan University, said: "In the field of structural biology, the analysis of the flexible structure of biological macromolecules is a long-term goal. The new algorithm developed by the Fudan team allows researchers to see key structural details through cryo-electron microscopy, which was not possible with previous techniques and will have important implications for biology, chemical research and drug discovery. ”
OPUS-DSD reconstruction structure model compared with traditional cryo-EM software parsing model: In the area marked by dotted lines, OPUS-DSD reconstructed model (green) has a more complete electron density than the traditional cryo-EM software parsed model (fuchsia red). This is because OPUS-DSD can reconstruct different 3D conformations separately without overlapping them in the same 3D model.
According to reports, with the rise of protein structure prediction technologies such as AlphaFold2 (AlphaFold2), computational biology has developed rapidly in recent years. Artificial intelligence technologies such as deep learning have become powerful assistants in structural biology experimental research, revealing the mysteries of life together with cryo-electron microscopy. In September this year, the Lasker Prize for Basic Medical Research, known as the "Nobel Prize weather vane", was announced, and the winners were DeepMind's Demis Hassabis and John Chop, who are the main inventors of AlphaFold2.
However, Professor Ma Jianpeng pointed out that in the field of structural biology, computer prediction technology is far from replacing traditional experimental structure measurement technology, and can only play a complementary and enhancing role. The structure of most biological macromolecules, especially those of very large complexes, will continue to be determined experimentally. In the process of experimental determination using cryo-electron microscopy, computer software plays an important role - by interpreting experimental measurement results, scientists can obtain more accurate images of biological structures.
Researchers at Fudan University operate cryo-electron microscopy.
Therefore, the algorithm of this type of software is crucial. Dr. Luo Zhenwei of Fudan University introduced that many functions of biological macromolecules are realized through their flexible properties, which are the main factors affecting the accuracy of structure determination. On the other hand, the signal-to-noise ratio of cryo-EM experimental data is very low, which brings great difficulties to the application of deep learning algorithms in this field. How to overcome the flexibility of biological macromolecular structures in cryo-EM data, especially the structural determination error caused by the flexibility of super-large complexes, is the focus and difficulty of current structural biology research.
The intelligent algorithm developed by the Multi-scale Research Institute of Complex Systems of Fudan University has overcome this world problem. The algorithm based on deep learning published in Nature Methods can effectively identify and process the flexible information of biological macromolecules, thereby improving the analysis ability of cryo-EM and obtaining the dynamic change information of the three-dimensional structure of biological macromolecules.
Conformational variations of OPUS-DSD resolution: Green and bronze represent two different conformations of OPUS-DSD resolution, respectively. In the area marked by the red dotted box, RNA strands in different conformations are in different positions, indicating that RNA strands are in dynamic motion. This dynamic structure information is difficult to extract and distinguish using traditional methods.
"This is an important achievement in the field of computational biology," said Yin Weihai, professor at the School of Biomedical Engineering at Shanghai Jiao Tong University and vice president of the Med-X Research Institute, "It not only has a profound impact on cryo-EM biomacromolecular structure elucidation technology, but also demonstrates the strength of the Fudan team to independently develop internationally leading algorithm software." In today's situation of limited procurement of computer hardware equipment, this achievement has the importance of 'insufficient computing power, algorithms to make up'. ”
With the publication of the paper, the OPUS-DSD algorithm has been open sourced on GitHub. In the future, the Fudan team will continue to use artificial intelligence as the technical hub to build a new generation of biological system analysis tools and methods. Ma Jianpeng said that these new tools and methods will help scientists interpret biological genetic information, support drug research and development through the prediction and design of the functional structure of biological macromolecules such as proteins and nucleic acids, and create an advanced technology platform for full-chain AI-empowered new drug research and development.
Editor-in-chief: Huang Haihua
Source: Author: Yu Taoran