During the past decade, due to the number of proteins in PDB database being increased gradually, traditional methods cannot better understand the function of newly discovered enzymes in chemical reactions. Computational models and protein feature representation for predicting enzymatic function are more important. Most of existing methods for predicting enzymatic function have used protein geometric structure or protein sequence alone. In this paper, the functions of enzymes are predicted from many-sided biological information including sequence information and structure information. Firstly, we extract the mutation information from amino acids sequence by the position scoring matrix and express structure information with amino acids distance and angle. Then, we use histogram to show the extracted sequence and structural features respectively. Meanwhile, we establish a network model of three parallel Deep Convolutional Neural Networks (DCNN) to learn three features of enzyme for function prediction simultaneously, and the outputs are fused through two different architectures. Finally, The proposed model was investigated on a large dataset of 43,843 enzymes from the PDB and achieved 92.34% correct classification when sequence information is considered, demonstrating an improvement compared with the previous result.
Cavity analysis in molecular dynamics is important for understanding molecular function. However, analyzing the dynamic pattern of molecular cavities remains a difficult task. In this paper, we propose a novel method to topologically represent molecular cavities by vectorization. First, a characterization of cavities is established through Word2Vec model, based on an analogy between the cavities and natural language processing (NLP) terms. Then, we use some techniques such as dimension reduction and clustering to conduct an exploratory analysis of the vectorized molecular cavity. On a real data set, we demonstrate that our approach is applicable to maintain the topological characteristics of the cavity and can find the change patterns from a large number of cavities.
Traditional biological experimental methods for enzyme function prediction have not been able to meet the increasing number of newly discovered enzymes measured by X-ray crystallography or magnetic resonance. A good computational model and protein feature representation for predicting enzymatic function can quickly annotate the functions of enzymes in chemical reactions. Existing machine learning methods usually compress protein 3D structure information into pictures convenient for convolutional neural networks (CNNs) and discard a large amount of relation information. Therefore, we proposed a method using the relation between amino acids directly to predict enzyme function. First, in addition to common structural features, we introduced a new structural feature, the relative angle of the amino acid (C–Cα–C) plane. Additionally, all protein structure features were organized into a new representation. Then, a structure relation network (SRN) to learn four features of the enzyme was established. Finally, the proposed model was evaluated on a large dataset containing 42,699 enzymes and achieved 92.08% classification accuracy, showing improvements compared with previous works.