Transfer learning via multi-scale convolutional neural layers for human–virus protein–protein interaction prediction

Abstract

Motivation To complement experimental efforts, machine learning-based computational methods are playing an increasingly important role to predict human-virus protein-protein interactions (PPIs). Furthermore, transfer learning can effectively apply prior knowledge obtained from a large source dataset/task to a small target dataset/task, improving prediction performance. Results To predict interactions between human and viral proteins, we combine evolutionary sequence profile features with a Siamese convolutional neural network (CNN) architecture and a multi-layer perceptron. Our architecture outperforms various feature encodings-based machine learning and state-of-the-art prediction methods. As our main contribution, we introduce two transfer learning methods (i.e., ‘frozen’ type and ‘fine-tuning’ type) that reliably predict interactions in a target human-virus domain based on training in a source human-virus domain, by retraining CNN layers. Finally, we utilize the ‘frozen’ type transfer learning approach to predict human-SARS-CoV-2 PPIs, indicating that our predictions are topologically and functionally similar to experimentally known interactions. Supplementary information Supplementary data are available at Bioinformatics online.

Cross-attention PHV: Prediction of human and virus protein-protein interactions using cross-attention–based neural networks

“…We compared the performance of cross-attention PHV with seven state-of-the-art methods, including Denovo [11], Zhou et al’s SVM-based method [14], Alguwaizani et al’s SVM-based method [15], Yang et al’s random forest–based and Doc2vec-based method [24], DeepViral [17], and Yang et al’s CNN-based method [16], using Denovo’s test dataset. As shown in Table 3, cross-attention PHV predicted the PPIs with an AC value >0.95 and outperformed the state-of-the-art models in five metrics, including SN, AC, AUC, MCC, and F1, demonstrating the superiority of cross-attention PHV.…”

“…Deep learning–based models have overcome such problems, however. For example, Yang et al embedded local features such as binding motifs into feature matrices and captured their patterns using a convolutional neural network (CNN) [16]. They applied two different transfer learning methods to improve the generalizability of the model.…”

Deep learning frameworks for protein–protein interaction prediction

“…PSSM is generated by applying Position-Specific Iterative (PSI)-BLAST searching in the protein database (like the UniRef50 database [54] ). In DPPI [33] and TransPPI [55] , the PSSM is a matrix , where is the length of the protein sequence and each element in the matrix denotes the probability of the amino acid in the position of the sequence. The only drawback of this method is that it needs an enormous effort for PSI-BLAST searching.…”

“…This approach [55] employs four connected convolutional layers followed with the pooling layers within a Siamese-like architecture to capture the latent patterns in the input protein sequence. The prediction module concatenates a pair of protein representations generated from two identical sub-networks and passes them through three stacked fully-connected layers followed with the leakyReLU activation.…”

Cross-attention PHV: Prediction of human and virus protein-protein interactions using cross-attention–based neural networks

“…Deep learning–based models have overcome such problems. For example, Yang et al embedded local features such as binding motifs into feature matrices and captured their patterns using a convolutional neural network (CNN) [19] . They applied two different transfer learning methods to improve the generalizability of the model.…”