WorkshopsStatistical Methods for cryo-EM Image Analysis
89
reads
In the past decades, cryo-electron microscopy (cryo-EM) has emerged as a powerful alternative to X-ray crystallography for structure determination of biological macro-molecules because it does not need crystals to solve the 3-D structure of the molecules. However, the cryo-EM data contain large number of 2-D projection images of single particles with extremely low signal-to-noise ratio and free orientations, which give many misaligned images as outliers, and a large number of clusters. Clustering analysis is a necessary step to group similar orientation projections for noise reduction while dimension reduction makes the computations for clustering analysis possible. In this talk, two statistical methods will be presented: multilinear principal component analysis (MPCA) for dimension reduction and -SUP algorithm for clustering analysis. Multilinear principal component analysis (MPCA) searches for low-dimensional projections on multilinear spaces simultaneously and, thereby, decreases dimensionality more effciently. -SUP is proved to be an optimum solution under the minimum -divergence criterion for mixture models of q-Gaussian distribution. Our simulation and real data analysis both show that it outperforms other clustering algorithms commonly used in the cryo-EM community. The collaborators of these works include Su-Yun Huang, Wei-hau Chang, Ting-Li Chen, Hung Hung, Dai-Ni Hsieh and Pei-Shien Wu.
reads
I-Ping Tu
2013-07-02
11:10:00 - 12:00:00
101 , Mathematics Research Center Building (ori. New Math. Bldg.)
In the past decades, cryo-electron microscopy (cryo-EM) has emerged as a powerful alternative to X-ray crystallography for structure determination of biological macro-molecules because it does not need crystals to solve the 3-D structure of the molecules. However, the cryo-EM data contain large number of 2-D projection images of single particles with extremely low signal-to-noise ratio and free orientations, which give many misaligned images as outliers, and a large number of clusters. Clustering analysis is a necessary step to group similar orientation projections for noise reduction while dimension reduction makes the computations for clustering analysis possible. In this talk, two statistical methods will be presented: multilinear principal component analysis (MPCA) for dimension reduction and -SUP algorithm for clustering analysis. Multilinear principal component analysis (MPCA) searches for low-dimensional projections on multilinear spaces simultaneously and, thereby, decreases dimensionality more effciently. -SUP is proved to be an optimum solution under the minimum -divergence criterion for mixture models of q-Gaussian distribution. Our simulation and real data analysis both show that it outperforms other clustering algorithms commonly used in the cryo-EM community. The collaborators of these works include Su-Yun Huang, Wei-hau Chang, Ting-Li Chen, Hung Hung, Dai-Ni Hsieh and Pei-Shien Wu.