DeepCF-PPI: improved prediction of protein-protein interactions by combining learned and handcrafted features based on attention mechanisms

Protein pairs can interact with each other by physical connections through electrostatic forces or hydrophobic effects. Prediction of protein-protein interactions (PPIs) is critical in many biological applications, including protein function identification, drug design, and disease detection. In thi...

Full description

Saved in:
Bibliographic Details
Published in:Applied intelligence (Dordrecht, Netherlands) Netherlands), 2023-07, Vol.53 (14), p.17887-17902
Main Authors: Tran, Hoai-Nhan, Xuan, Quynh Nguyen Phuc, Nguyen, Tuong-Tri
Format: Article
Language:English
Subjects:
Accuracy
Amino acids
Artificial Intelligence
Artificial neural networks
Composition
Computer architecture
Computer Science
Datasets
Machines
Manufacturing
Mechanical Engineering
Natural language processing
Processes
Proteins
Source code
Yeast
Citations: Items that this one cites
Items that cite this one
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Protein pairs can interact with each other by physical connections through electrostatic forces or hydrophobic effects. Prediction of protein-protein interactions (PPIs) is critical in many biological applications, including protein function identification, drug design, and disease detection. In this paper, we propose a method of combining features for protein representation and PPI prediction directly from protein sequences. First, we utilized 5 protein sequence extractors including amino acid composition, pseudo-amino acid composition, amphiphilic pseudo-amino acid composition, quasi-sequence-order, and dipeptide composition to extract handcrafted features. Next, we applied a natural language processing technique, Word2vec, to generate learned features by embedding protein sequences into a feature space. Finally, a deep neural network architecture was employed for combining two types of features and identifying PPIs. The proposed method was evaluated on the Yeast core, Human, and eight independent datasets. The experimental results show that our method achieved 95.6% of accuracy on Yeast core, 99.2% of accuracy on Human, and 100% of accuracy on eight independent datasets, respectively. In addition, we also made extensive comparisons with other existing prediction methods, and the experimental results demonstrated the superior ability of our proposed method. The datasets and source code of our method can be downloaded at https://github.com/thnhan/DeepCF-PPI.git .
ISSN:0924-669X
1573-7497
DOI:10.1007/s10489-022-04387-2