DeepUbi: a deep learning framework for prediction of ubiquitination sites in proteins

Abstract

Background: Protein ubiquitination occurs when the ubiquitin protein binds to a target protein residue of lysine (K), and it is an important regulator of many cellular functions, such as signal transduction, cell division, and immune reactions, in eukaryotes. Experimental and clinical studies have shown that ubiquitination plays a key role in several human diseases, and recent advances in proteomic technology have spurred interest in identifying ubiquitination sites. However, most current computing tools for predicting target sites are based on small-scale data and shallow machine learning algorithms.

Results: As more experimentally validated ubiquitination sites emerge, we need to design a predictor that can identify lysine ubiquitination sites in large-scale proteome data. In this work, we propose a deep learning predictor, DeepUbi, based on convolutional neural networks. Four different features are adopted from the sequences and physicochemical properties. In a 10-fold cross validation, DeepUbi obtains an AUC (area under the Receiver Operating Characteristic curve) of 0.9, and the accuracy, sensitivity and specificity exceeded 85%. The more comprehensive indicator, MCC, reaches 0.78. We also develop a software package that can be freely downloaded from https://github.com/Sunmile/DeepUbi .

Conclusion: Our results show that DeepUbi has excellent performance in predicting ubiquitination based on large data.

Keywords: Convolutional neural networks; Deep learning; Ubiquitination.

Fig. 1

ROC curves of different cross-validations. ROC curves and their AUC values of 4-, 6-, 8-, and 10-fold cross validations with the One-Hot encoding scheme

Fig. 2

ROC curves of different feature constructions. ROC curves and their AUC values of four features in the 10-fold cross validation. These curves are very close to each other which illustrate the robustness of the model

Fig. 3

Different sequence analysis charts about ubiquitination and non-ubiquitination peptides. a A bar chart to compare the number of flanking amino acids surrounding the ubiquitination and non-ubiquitination peptides. b A circular chart to compare the percentage of flanking amino acids surrounding the ubiquitination and non-ubiquitination peptides. c Two Sample Logos web-server to calculate and visualize differences between ubiquitination and non-ubiquitination peptides

Fig. 4

Flow chart of the data collection and processing. Firstly, collecting the raw proteins and then removing the redundant protein sequences with CD-Hit; secondly, intercepting the protein sequences with a 31 sliding window to get the positive and negative fragments; at last, using 30% identity in negative samples to get a balanced training data

Fig. 5

a Flow chart of the CNN deep learning model. b An example of convolution-pooling structure. a Input a fragment and encode; construct an embedding layer; build multi-convolution-pooling layers; construct fully connected layers; and then get the output. b Use different filters with different sizes to get a series of feature maps; and then use a max-pooling and concatenating together to form a feature vector. Finally, the softmax function regularization is used to get the classification