蛋白质冷冻电镜图像的分类算法本科论文.doc

PAGE \* MERGEFORMAT PAGE \* MERGEFORMAT VII PAGE \* MERGEFORMAT PAGE \* MERGEFORMAT I 论文 蛋白质冷冻电镜图像的分类算法研究 Classification of Protein Cryo-EM Images 蛋白质冷冻电镜图像的分类算法研究 摘要 本篇论文的主要工作是对模拟的蛋白质二维投影图像进行分类。这个课题源于冷冻电镜蛋白质的三维重构，对蛋白质投影图像的分类平均是蛋白质三维重构的重要前提。我们首先根据蛋白质冷冻电镜图像的形成原理，模拟产生不同投影角度的蛋白质二维图像。然后提取图像的平移旋转不变量，为之后的分类做准备。由于模拟产生的图像中蛋白质的投影图像都是居中的，所以可近似忽略其平移变量。旋转不变量是通过对图像做极坐标变换和傅里叶变换而提取到的。接着运用k-medoids算法,一种典型的基于划分的聚类方法,对处理后的图像进行分类。最后采用了由Fowlkes和Mallows提出的一种评估聚类结果的方法，用来验证本课题的分类结果。在结果分析中，我们给输入图像分别加入不同的高斯白噪声，然后对比不同信噪比图片的聚类结果，得出该算法能够处理的信噪比的范围。此外，我们还对比了在输入图像总数不同或类别数不同时的分类结果。以此来全面评估k-medoids算法对蛋白质投影图像的分类效果。 关键词：蛋白质，图像，分类，旋转不变性，k-medoids算法 CLASSIFICATION OF PROTEIN CRYO-EM IMAGES Abstract In this paper our main work is to classify simulated protein two-dimensional projection images. This problem is related with three-dimensional reconstruction of protein structure from cryo-EM data, the process of classification and averaging of the projected image is an important prerequisite for three-dimensional reconstruction of protein. First, we produce simulated projection images of protein from various directions, in accordance with the actual collection of cryo-EM images . Then extract the translation and rotation invariant from these images, to be used as features for the following classification. The projected images are centered when we produce simulation images, so it can be approximated by ignoring its translational variables. We use the polar coordinate transformation and Fourier transform to obtain rotational invariances of images. Next use classical clustering methods, namely k-medoids algorithm, to classify the images. Finally, in order to assess the quality of the classification results, we use the index proposed by the Fowlkes and Mallows. In the experiments, we added different Gaussian white noise to the input image, and compared the results of clustering with different signal to noise ratio,. In order to fully assess the k-medoids algorithm for projected image classificatio