Deep Neural Network Based Predictions of Protein Interactions Using Primary Sequences

Abstract

Machine learning based predictions of protein–protein interactions (PPIs) could provide valuable insights into protein functions, disease occurrence, and therapy design on a large scale. The intensive feature engineering in most of these methods makes the prediction task more tedious and trivial. The emerging deep learning technology enabling automatic feature engineering is gaining great success in various fields. However, the over-fitting and generalization of its models are not yet well investigated in most scenarios. Here, we present a deep neural network framework (DNN-PPI) for predicting PPIs using features learned automatically only from protein primary sequences. Within the framework, the sequences of two interacting proteins are sequentially fed into the encoding, embedding, convolution neural network (CNN), and long short-term memory (LSTM) neural network layers. Then, a concatenated vector of the two outputs from the previous layer is wired as the input of the fully connected neural network. Finally, the Adam optimizer is applied to learn the network weights in a back-propagation fashion. The different types of features, including semantic associations between amino acids, position-related sequence segments (motif), and their long- and short-term dependencies, are captured in the embedding, CNN and LSTM layers, respectively. When the model was trained on Pan’s human PPI dataset, it achieved a prediction accuracy of 98.78% at the Matthew’s correlation coefficient (MCC) of 97.57%. The prediction accuracies for six external datasets ranged from 92.80% to 97.89%, making them superior to those achieved with previous methods. When performed on Escherichia coli, Drosophila, and Caenorhabditis elegans datasets, DNN-PPI obtained prediction accuracies of 95.949%, 98.389%, and 98.669%, respectively. The performances in cross-species testing among the four species above coincided in their evolutionary distances. However, when testing Mus Musculus using the models from those species, they all obtained prediction accuracies of over 92.43%, which is difficult to achieve and worthy of note for further study. These results suggest that DNN-PPI has remarkable generalization and is a promising tool for identifying protein interactions.

Sequence representations and their utility for predicting protein-protein interactions

“…These methods have included the Support Vector Machine [15], Random Forests [29] and autoencoders [6]. More recently, studies such as DPPI [31], DNN-PPI [32], and PIPR [33] have explored deep learning frameworks for PPI prediction. Note that these newer deep learning-based approaches are end-to-end classification models and do not specifically focus on the feature construction technique.…”

“…For the human dataset, it is not the best among the machine learning based methods. In these experiments, we compare the results of our methods (M9 and M10) on Guo's yeast dataset with baseline approaches SVM-AC [15], kNN-CTD [17], EELM-PCA [46], SVM-MCD [27], MLP [47], RF-LPQ [28], SAE [6], DNN-PPI [32], DPPI [31] and SRGRU, SCNN and PIPR from [33]. The results for the baseline approaches are taken from [33].…”

NeuRiPP: Neural network identification of RiPP precursor peptides

“…Neural networks are also gaining popularity in their application to biological systems. Some examples of these in the context of biological sequences are DNNs trained to identify lab origin given a DNA sequence (15), identify whether a sequence of DNA is plasmid or chromosomal in origin (16), and predicting protein-protein interactions between two proteins (17). In this study, I explore whether DNN architectures successful in image and text classifiers are suitable for the problem of identifying putative PPs.…”

Mutation effect estimation on protein-protein interactions using deep contextualized representation learning

“…Recently, deep learning methods show the potential of automatically extracting comprehensive features from protein sequences, and gain unprecedented success in PPI tasks. Corresponding methods, including DNN-PPI [28], DPPI [18], and PIPR [8] employ various neural sequence pair models to predict PPI information based on protein sequences. In the application of predicting the structural properties of a protein complex, NetSurfp-2.0 [25] employs a neural sequence model to predict the solvent accessible surface area (ASA) of a protein.…”

“…Such mechanisms deploy fixed representations, e.g. one-hot vectors [28], physicochemical property-aware encoding [8] or static amino acid embeddings [8]. However, these representations face several indispensable problems for PPI property predictions upon mutations.…”