基于卷积神经网络的蛋白质折叠类型最小特征提取

Extraction of minimal representation of protein folds based on convolutional neural network

通过蛋白质的序列、结构等信息构建完整的蛋白质宇宙是生物信息学中的重要课题,相关研究对蛋白质结构预测、蛋白质进化路径分析以及蛋白质结构设计等方面的研究都有重要的意义.从蛋白质结构的一种简化表示——蛋白质接触图出发,通过训练卷积神经网络进行特征提取,筛选出可识别结构域折叠类型的最小特征向量,构建蛋白质折叠类型空间,并使用谱聚类等方法对不同蛋白质折叠类型的高维分布情况进行分析.得到的最小特征向量兼顾了信息的完整性与冗余度,可以很好地表示全部七种常见蛋白质类的空间关联.该研究结果填补了之前蛋白质宇宙研究中对不常见类的空间位置和相互关系描述的空白,加深了对于蛋白质结构相似性的理解.

Establishing an entire protein universe from sequential and structural information is a key problem in bioinformatics,and is of great importance in protein structure prediction,protein evolution analysis and protein structure design. In this paper,starting from a simplified representaion of protein structure,contact map,we trained a deep convolutional neural network(DCNN) and studied the shortest feature vectors that were able to recognize different protein folds correctly. We constructed a space of protein folds spanned with these shortest feature vectors,and analyzed the highdimensional distribution with spectral clustering and other methods. With these shortest feature vectors, both information integrity and redundance are considered and all the seven common protein classes and their spatial relationships can be characterized. Our research fills gaps in the description of spatial position and relationship of classes which is absent from previous researches and may improve the understanding on similarity between protein classes.