Structure analysis of biological macromolecules is one of the key technologies in biological science, and cryo-electron microscopy (cryo-EM) is one of the research tools for obtaining high-resolution 3D structures.
Recently, Dr Jianpeng Ma, Director of Multiscale Research Institute for Complex Systems, Fudan University, Adjunct Professor of Shanghai Institute for Advanced Study, Zhejiang University, developed an advanced intelligent algorithm OPUS-DSD and published relevant work as ‘OPUS-DSD: deep structural disentanglement for cryo-EM single-particle analysis’ in Nature Methods.
The algorithm successfully analyzed biomolecule structures e.g., proteins, nucleic acids, protein/nucleic acid complexes etc that could not be solved by traditional techniques, and demonstrated efficiency and accuracy in distinguishing conformation distribution of flexible domains in sample structures. Such new method could effectively establish high-precision structural models, thus preventing new drug development failures from inaccurate structures of target proteins.
Flexibility of bimolecular structure is essential to its function, but it is one of the main contributors to inaccuracy in structure prediction. Conformational diversity caused by structural flexibility in cryo-EM data processing makes it challenging to obtain accurate 3D models from individual samples, and at the same time, signal-to-noise ratio from experimental data tends to be extremely low, posing difficulty in applying deep learning algorithms in this field. Thus, overcoming the error in structure prediction caused by flexibility of biomolecule structures in cryo-EM data, especially flexibility of ultra-large complexes, is the bottleneck for current structural biology study.
Professor Ma and his team introduced a deep learning-based computational method to effectively recognize and process biomacromolecule flexibility to obtain information on dynamical changes of cryo-EM three-dimensional structures. The results were demonstrated as below.
Comparison of structured re-constructed by OPUS-DSD and traditional cryo-EM modeling software. Density map of structured re-constructed by OPUS-DSD (in green) is more digitally complete compared with that of traditional cryo-EM software (in purple) due to the algorithm's capability to distinguishing 3D conformation instead of overlapping them in one 3D structure (dotted area). Heterogeneity analysis of the Sc80S ribosome (EMPIAR: 10002). Both maps are contoured at the same level and visualized by ChimeraX.
Conformational change reconstructed by OPUS-DSD, presented by green and bronze respectively. Within dotted area RNA are in different position, showing the RNA in dynamical movement. Such conformational structure could not be exacted by traditional techniques.
OPUS-DSD demonstrated superiority in data processing and robustness in maintaining high accuracy in data with lower signal-to-noise ratio. In addition, it could be expanded from single-particle cryo-EM to highend and lower signal-to-noise ratio cryo-tomography (Cryo-ET) studies.
The team will continue to leverage AI technology to build tools and methods for analyzing biological systems to decipher biogenetic information, and to facilitate drug discovery and development by predicting and designing the functional structures of biological macromolecules such as proteins and nucleic acids, and to establish advanced technology platform for AI-enabled new drug discovery and development.
The work is available at https://www.nature.com/articles/s41467-023-41489-y and the open-source algorithm could be accessed at https://github.com/alncat/opusDSD.