Xu Lab Publishes in PNAS
Unsupervised 3D detection of complex macromolecular structures using Cryo-ET
Cryo-electron tomography (cryo-ET) represents a formidable tool for the direct visualization of intricate macromolecular structures in their natural cellular milieu. The automated detection of such structures is a critical prerequisite for comprehending protein function and interactions within cells.
Current computational methods have relied upon either known structural templates or manually annotated training datasets, thus impeding further applications of detection for unknown structures.
To address this limitation, a recent study published in Proceedings of the National Academy of Sciences spearheaded by Prof. Min Xu's lab has put forth a novel approach named the Deep Iterative Subtomogram Clustering Approach (DISCA). This approach employs deep learning techniques to recognize subsets of homogeneous structures, without relying on templates or training datasets. Specifically, DISCA capitalizes on 3D structural features and their distributions to model the heterogeneity of macromolecular complexes.
As noted in the paper, "DISCA successfully detects a diverse range of structures with varying molecular sizes," making it an unbiased and systematic approach for the recognition of macromolecular complexes in situ”, according to Prof. Xu.
Cryoelectron tomography directly visualizes heterogeneous macromolecular structures in their native and complex cellular environments. The automatic detection of macromolecular complexes is an open and challenging problem in cellular cryoelectron tomography. Existing computational methods rely on known structural templates or manually labeled training datasets. In this paper, members of Xu Lab introduce a high-throughput template-and-label-free deep learning approach, Deep Iterative Subtomogram Clustering Approach (DISCA), that automatically detects subsets of homogeneous structures by learning and modeling 3D structural features and their distributions. Evaluation of five experimental cryo-ET datasets shows that an unsupervised deep learning-based method can detect diverse structures with a wide range of molecular sizes. This unsupervised detection is an important step towards systematic unbiased recognition of macromolecular complexes in situ.
X. Zeng, A. Kahng, L. Xue, J. Mahamid, Y.W. Chang and M. Xu, 2023. High-throughput cryo-ET structural pattern mining by deep iterative unsupervised clustering. Proceedings of the National Academy of Sciences, 120 (15) e2213149120.