Fer-COCL: A Novel Method Based on Multiple Deep Learning Algorithms for Identifying Fertility-Related Proteins

Abstract

The survival of species depends on the fertility of organisms. It is also worthwhile to study the proteins that can regulate the reproductive activity of organisms. Since biological experiments are laborious to confirm proteins, it has become a priority that develop relevant computational models to predict the function of fertility-related proteins. With the development of machine learning, pertinent various algorithms can be the key to identifying fertility-related proteins. In this work, we develop a model Fer-COCL based on deep learning. The model consists of multiple features as well as multiple deep learning algorithms. First, we extract features using Amino acid composition (AAC), Dipeptide composition (DPC), CTD transition (CTDT) and deviation between the dipeptide and the expected mean (DDE). After that, the spliced features are fed into the classifier. The data processed jointly by convolutional neural network and long short-term memory is input to the fully connected layer for classification. After evaluating the model using 10-fold cross-validation, the accuracy of the two data sets reaches 97.1% and 98.3%, respectively. The results indicate that the model is efficient and accurate, facilitating biologists' research on biological fertility. In addition, a free online tool for predicting the function of fertility-related proteins is available at http://fercocl.zhanglab.site/.