CoChat
Sign Up
Cited by 10Machine Learning in BioinformaticsBioinformatics and Genomic NetworksProtein Structure and Dynamics

The DOI matches its Crossref record; 3 of the 3 references it cites were found in a registry or an index.

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

Open access
Tsukiyama, Sho, Kurata, Hiroyuki·2022·Computational and Structural Biotechnology Journal
DOI 10.1016/j.csbj.2022.10.012PMC PMC9546503

Published version of a preprint

Open this paper in CoChat — free accountOpen PDF

Abstract

Viral infections represent a major health concern worldwide. The alarming rate at which SARS-CoV-2 spreads, for example, led to a worldwide pandemic. Viruses incorporate genetic material into the host genome to hijack host cell functions such as the cell cycle and apoptosis. In these viral processes, protein-protein interactions (PPIs) play critical roles. Therefore, the identification of PPIs between humans and viruses is crucial for understanding the infection mechanism and host immune responses to viral infections and for discovering effective drugs. Experimental methods including mass spectrometry-based proteomics and yeast two-hybrid assays are widely used to identify human-virus PPIs, but these experimental methods are time-consuming, expensive, and laborious. To overcome this problem, we developed a novel computational predictor, named cross-attention PHV, by implementing two key technologies of the cross-attention mechanism and a one-dimensional convolutional neural network (1D-CNN). The cross-attention mechanisms were very effective in enhancing prediction and generalization abilities. Application of 1D-CNN to the word2vec-generated feature matrices reduced computational costs, thus extending the allowable length of protein sequences to 9000 amino acid residues. Cross-attention PHV outperformed existing state-of-the-art models using a benchmark dataset and accurately predicted PPIs for unknown viruses. Cross-attention PHV also predicted human-SARS-CoV-2 PPIs with area under the curve values >0.95. The Cross-attention PHV web server and source codes are freely available at https://kurata35.bio.kyutech.ac.jp/Cross-attention_PHV/ and https://github.com/kuratahiroyuki/Cross-Attention_PHV, respectively.

Overview

The claims and statistics of this paper haven’t been checked yet.

Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.

A few free checks a day · papers already verified open instantly · results are public

Integrity signals

Registry

Crossref · 10.1016/j.csbj.2022.10.012

Cites

3 of 3 references verified0 citation errors · 0 not checkable

Statistics

Figures

Graphical abstract
Graphical abstract CC BY · PubMed Central
Fig. 1 Workflow of word2vec-based encoding. (A) Amino acid sequences were converted into arrangements of consecutive 4-mers. (B) Amino acid sequences in the UniProtKB/Swiss-Prot database were converted as representations of 4-mers and used for training the word2vec model. (C) Each 4-mer in the amino acid sequence was converted into a feature vector using the trained word2vec model. The resultant feature vectors were concatenated into a feature matrix.
Fig. 1 Workflow of word2vec-based encoding. (A) Amino acid sequences were converted into arrangements of consecutive 4-mers. (B) Amino acid sequences in the UniProtKB/Swiss-Prot database were converted as representations of 4-mers and used for training the word2vec model. (C) Each 4-mer in the amino acid sequence was converted into a feature vector using the trained word2vec model. The resultant feature vectors were concatenated into a feature matrix.CC BY · PubMed Central
Fig. 2 Structure of cross-attention PHV. Cross-attention PHV is composed of three sub-networks. The word2vec (W2V)-based feature matrices of humans and viruses were input into the convolutional embedding module. To extract interaction features between two protein sequences, multi-head attention layers were employed in the cross-attention module. Finally, the feature vectors generated by the global max-pooling layer were concatenated to compute a final score through three linear layers.
Fig. 2 Structure of cross-attention PHV. Cross-attention PHV is composed of three sub-networks. The word2vec (W2V)-based feature matrices of humans and viruses were input into the convolutional embedding module. To extract interaction features between two protein sequences, multi-head attention layers were employed in the cross-attention module. Finally, the feature vectors generated by the global max-pooling layer were concatenated to compute a final score through three linear layers.CC BY · PubMed Central
Fig. 3 Prediction performance of word2vec-based cross-attention PHV with respect to k-mer value. The models were evaluated via 5-fold cross-validation on Denovo's training dataset.
Fig. 3 Prediction performance of word2vec-based cross-attention PHV with respect to k-mer value. The models were evaluated via 5-fold cross-validation on Denovo's training dataset.CC BY · PubMed Central
Fig. 4 Comparison of performance between the word2vec-based and binary encodings in cross-attention PHV. Models trained via 5-fold cross-validation were evaluated with Denovo's test dataset.
Fig. 4 Comparison of performance between the word2vec-based and binary encodings in cross-attention PHV. Models trained via 5-fold cross-validation were evaluated with Denovo's test dataset.CC BY · PubMed Central
Fig. 5 Comparison of performance between cross-attention–based and self-attention–based neural networks. Models trained via 5-fold cross-validation were evaluated with Denovo's test dataset.
Fig. 5 Comparison of performance between cross-attention–based and self-attention–based neural networks. Models trained via 5-fold cross-validation were evaluated with Denovo's test dataset.CC BY · PubMed Central
Fig. 6 Comparison of the performance of cross-attention PHV and LSTM-PHV in predicting PPIs for unknown viruses. (A) Performance on the H1N1 dataset, which regards H1N1 as an unknown virus. (B) Performance on the H3N2 dataset, which regards H3N2 as an unknown virus. (C) Performance on the H5N1 dataset, which regards H5N1 as an unknown virus.
Fig. 6 Comparison of the performance of cross-attention PHV and LSTM-PHV in predicting PPIs for unknown viruses. (A) Performance on the H1N1 dataset, which regards H1N1 as an unknown virus. (B) Performance on the H3N2 dataset, which regards H3N2 as an unknown virus. (C) Performance on the H5N1 dataset, which regards H5N1 as an unknown virus.CC BY · PubMed Central
Fig. 7 Comparison of the performance of cross-attention PHV with LSTM-PHV in predicting human–SARS-CoV-2 PPIs. (A) Performance on a balanced dataset (positive:negative = 1:1). (B) Performance on an imbalanced dataset (positive:negative = 1:5).
Fig. 7 Comparison of the performance of cross-attention PHV with LSTM-PHV in predicting human–SARS-CoV-2 PPIs. (A) Performance on a balanced dataset (positive:negative = 1:1). (B) Performance on an imbalanced dataset (positive:negative = 1:5).CC BY · PubMed Central
Fig. 8 t-SNE–based visualization of features generated during prediction of PPIs using the HuV-PPI test datasets. The word2vec-based feature matrices, hidden feature matrices, and feature vectors were retrieved from the neural networks. The feature matrices were transformed into vectors by sampling the maximum values of each feature. The human and virus feature vectors were then concatenated. The t-SNE maps for the H1N1, H3N2, and H5N1 datasets are shown at the left, center, and right, respectively. Blue, yellow, green, and red marks indicate false-positive, false-negative, true-negative, and true-positive samples, respectively. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)
Fig. 8 t-SNE–based visualization of features generated during prediction of PPIs using the HuV-PPI test datasets. The word2vec-based feature matrices, hidden feature matrices, and feature vectors were retrieved from the neural networks. The feature matrices were transformed into vectors by sampling the maximum values of each feature. The human and virus feature vectors were then concatenated. The t-SNE maps for the H1N1, H3N2, and H5N1 datasets are shown at the left, center, and right, respectively. Blue, yellow, green, and red marks indicate false-positive, false-negative, true-negative, and true-positive samples, respectively. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)CC BY · PubMed Central
Fig. 9 t-SNE–visualized map of the respective human and virus feature vectors on the HuV-PPI datasets.
Fig. 9 t-SNE–visualized map of the respective human and virus feature vectors on the HuV-PPI datasets.CC BY · PubMed Central

Verification

Claims the paper makes

The claims and statistics of this paper haven’t been checked yet.

Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.

A few free checks a day · papers already verified open instantly · results are public

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

The self-check needs the paper’s full text.

Citations

Cites

Research works printed but not linked: 40

  1. Burckhardt CJ Greber UF. Virus movements on the plasma membrane support infection and transmission between cells PLoS Pathogens 2009;5:e1000621-e1000621. · DOI
  2. 10.1083/jcb.201108131 · DOI
  3. 10.1016/j.jmb.2018.06.024 · DOI
  4. Parnell G., McLean A., Booth D. et al: Aberrant cell cycle and apoptotic changes characterise severe influenza A infection–a meta-analysis of genomic signatures in circulating leukocytes. PLoS ONE 2011; 6: e17186. · DOI
  5. Li F.Q., Tam J.P. and Liu D.X.: Cell cycle arrest and apoptosis induced by the coronavirus infectious bronchitis virus in the absence of p53. Virology 2007; 365: 435. · DOI
  6. Gioti K., Kottaridi C., Voyiatzaki C. et al: Animal coronaviruses induced apoptosis. Life 2021; 11. · DOI
  7. Mir A., Naghibzadeh M. and Saadati N.: INDEX: Incremental depth extension approach for protein–protein interaction networks alignment. Biosystems 2017; 162: 24. · DOI
  8. Hayashi T., Matsuzaki Y., Yanagisawa K. et al: MEGADOCK-Web: an integrated database of high-throughput structure-based protein-protein interaction predictions. BMC Bioinf 2018; 19: 62. · DOI
  9. Singh R Park D Xu J et al. Struct2Net: a web service to predict protein-protein interactions using a structure-based approach Nucleic Acids Res 2010;38:W508-515. · DOI
  10. Eid F.E., ElHefnawi M. and Heath L.S.: DeNovo: virus-host sequence-based protein-protein interaction prediction. Bioinformatics 2016; 32: 1144. · DOI
  11. Barman RK Saha S Das S. Prediction of interactions between viral and host proteins using supervised machine learning methods PLoS One 2014;9:e112034. · DOI
  12. Dey L., Chakraborty S. and Mukhopadhyay A.: Machine learning techniques for sequence-based prediction of viral-host interactions between SARS-CoV-2 and human proteins. Biomed J 2020; 43: 438. · DOI
  13. 10.3934/mbe.2020176 · DOI
  14. 10.1016/j.csbj.2022.06.025 · DOI
  15. Zhou X., Park B., Choi D. et al: A generalized approach to predicting protein-protein interactions between virus and host. BMC Genomics 2018; 19: 568. · DOI
  16. Alguwaizani S., Park B., Zhou X. et al: Predicting interactions between virus and host proteins using repeat patterns and composition of amino acids. J Healthcare Eng 2018; 2018: 1391265. · DOI
  17. Yang X., Yang S., Lian X. et al: Transfer learning via multi-scale convolutional neural layers for human–virus protein–protein interaction prediction. Bioinformatics 2021; 37: 4771. · DOI
  18. 10.1093/bioinformatics/btab147 · DOI
  19. Mikolov T Chen K Corrado G et al. Efficient estimation of word representations in vector space. 2013 arXiv:1301.3781. · arXiv
  20. Mikolov T Sutskever I Chen K et al. Distributed representations of words and phrases and their compositionality. 2013 arXiv:1310.4546. · arXiv
  21. Tsukiyama S Hasan MM Deng H-W et al. BERT6mA: prediction of DNA N6-methyladenine site using deep learning-based approaches Briefings in Bioinformatics 2022;23:bbac053. · DOI
  22. Charoenkwan P Nantasenamat C Hasan MM et al. BERT4Bitter: a bidirectional encoder representations from transformers (BERT)-based model for improving the prediction of bitter peptides Bioinformatics 2021:btab133. · DOI
  23. Kurata H., Tsukiyama S. and Manavalan B.: iACVP: markedly enhanced identification of anti-coronavirus peptides using a dataset-specific word2vec model. Briefings Bioinf 2022; 23. · DOI
  24. Zhang Y.F., Wang X., Kaushik A.C. et al: SPVec: A Word2vec-inspired feature representation method for drug-target interaction prediction. Front Chem 2019; 7: 895. · DOI
  25. 10.1016/j.csbj.2019.12.005 · DOI
  26. Tsukiyama S Hasan MM Fujii S et al. LSTM-PHV: prediction of human-virus protein–protein interactions by LSTM with word2vec Briefings in Bioinform 2021;22:bbab228. · DOI
  27. Chen X Cheng Y Wang S et al. EarlyBERT: Efficient BERT Training via Early-bird Lottery Tickets. 2020 arXiv:2101.00063. · arXiv
  28. Lei T. When Attention Meets Fast Recurrence: Training Language Models with Reduced Compute. 2021 arXiv:2102.12459. · DOI
  29. Mildenhall B Srinivasan PP Tancik M et al. NeRF: Representing scenes as neural radiance fields for view synthesis. 2020 arXiv:2003.08934. · DOI · Paper page
  30. Tolstikhin I Houlsby N Kolesnikov A et al. MLP-Mixer: An all-MLP Architecture for Vision. 2021 arXiv:2105.01601. · arXiv
  31. 10.1093/bib/bbaa425 · DOI
  32. The UniProt Consortium. UniProt: the universal protein knowledgebase Nucleic Acids Res 2017;45:D158-D169. · DOI · Paper page
  33. Henikoff S. and Henikoff J.G.: Amino acid substitution matrices from protein blocks. PNAS 1992; 89: 10915. · DOI · Paper page
  34. 10.1002/pro.3978 · DOI · Paper page
  35. 10.1093/bioinformatics/bts565 · DOI · Paper page
  36. Vaswani A Shazeer N Parmar N et al. Attention is all you need. 2017 arXiv:1706.03762. · arXiv
  37. Wu H Gu X. Max-pooling dropout for regularization of convolutional neural networks. 2015 arXiv:1512.01400. · DOI
  38. Pedregosa F Varoquaux G Gramfort A et al. Scikit-learn: machine learning in python. 2012 arXiv:1201.0490. · arXiv
  39. van der Maaten L. and Hinton G.: Visualizing data using t-SNE. J Machine Learn Res 2008; 9: 2579.
  40. Wu R Ding F Wang R et al. High-resolution de novo structure prediction from primary sequence bioRxiv 2022:2022.2007.2021.500999.

Metadata

Authors
Tsukiyama, S., & Kurata, H.
Affiliations
Kyushu Institute of Technology (JP)
Corresponding author
Kurata, Hiroyuki
ORCID
Venue
Computational and Structural Biotechnology Journal
Publisher
American Association for the Advancement of Science (AAAS)
Published
2022-01
Volume
20
Pages
5564-5573
ISSN
2001-0370 (ISSN-L)
DOI
10.1016/j.csbj.2022.10.012
PMC
PMC9546503
OpenAlex
W4303520619
Cited by
10
References
3
Citations by year
2026: 2 · 2025: 4 · 2024: 1 · 2023: 3
FWCI
0.65
Type
journal article
Language
en
Open access
Licence
Locations
  • Computational and Structural Biotechnology Journal (publishedVersion, cc-by)
  • bioRxiv (Cold Spring Harbor Laboratory) (acceptedVersion, cc-by-nc)
  • PubMed Central (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
Funders
Japan Society for the Promotion of Science (22H03688, 22J22706)
Topics
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Machine Learning in Bioinformatics
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Bioinformatics and Genomic Networks
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Protein Structure and Dynamics
Keywords
protein-protein interaction prediction, human-virus interactions, cross-attention, 1D convolutional neural network, SARS-CoV-2, viral infection, host-virus interaction, web server

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review