An interactome landscape of SARS-CoV-2 virus-human protein-protein interactions by protein sequence-based multi-label classifiers
Lee, H.-J.
Show abstract
The new coronavirus species, SARS-CoV-2, caused an unprecedented global pandemic of COVID-19 disease since late December 2019. A comprehensive characterization of protein-protein interactions (PPIs) between SARS-CoV-2 and human cells is a key to understanding the infection and preventing the disease. Here we present a novel approach to predict virus-host PPIs by multi-label machine learning classifiers of random forests and XGBoost using amino acid composition profiles of virus and human proteins. Our models harness a large-scale database of Viruses.STRING with >80,000 virus-host PPIs along with evidence scores for multi-level evidence prediction, which is distinct from predicting binary interactions in previous studies. Our multi-label classifiers are based on 5 evidence levels binned from evidence scores. Our best model of XGBoost achieves 74% AUC and 68% accuracy on average in 10-fold cross validation. The most important amino acids are cysteine and histidine. In addition, our model predicts experimental PPIs with higher accuracy than text mining-based PPIs by 4% despite their smaller data size by more than 6-fold. We then predict evidence levels of [~]2,000 SARS-CoV-2 virus-human PPIs from public experimental proteomics data. Interactions with SARS-CoV-2 Nsp7b show high evidence. We also predict evidence levels of all pairwise PPIs of [~]550,000 between the SARS-CoV-2 and human proteomes to provide a draft virus-host interactome landscape for SARS-CoV-2 infection in humans in a comprehensive and unbiased way in silico. Most human proteins from 140 highest evidence predictions interact with SARS-CoV-2 Nsp7, Nsp1, and ORF14, with significant enrichment in the top 2 pathways of vascular smooth muscle contraction (CALD1, NPR2, CALML3) and Myc targets (CBX3, PES1). Our prediction also suggests that histone H2A components are targeted by multiple SARS-CoV-2 proteins.
Matching journals
The top 4 journals account for 50% of the predicted probability mass.
Similar papers in this journal
- Transfer learning via multi-scale convolutional neural layers for human-virus protein-protein interaction prediction 98%
- FAPM: Functional Annotation of Proteins using Multi-Modal Models Beyond Structural Modeling 95%
- DualNetGO: A Dual Network Model for Protein Function Prediction via Effective Feature Selection 95%
Similar papers in this journal
- CoRegNet: Unraveling Gene Co-regulation Networks from Public RNA-Seq Repositories Using a Beta-Binomial Statistical Model 95%
- Structural Coverage of the Human Interactome 94%
- EGRET: Edge Aggregated Graph Attention Networks and Transfer Learning Improve Protein-Protein Interaction Site Prediction 94%
Similar papers in this journal
- Accurate Prediction of Virus-Host Protein-Protein Interactions via a Siamese Neural Network Using Deep Protein Sequence Embeddings 96%
- Discovering nuclear localization signal universe through a novel deep learning model with interpretable attention units 95%
- Network-based Virus-Host Interaction Prediction with Application to SARS-CoV-2 95%
Similar papers in this journal
- Paying Attention to Attention: High Attention Sites as Indicators of Protein Family and Function in Language Models 95%
- Zero-shot segmentation using embeddings from a protein language model identifies functional regions in the human proteome 95%
- Protein prediction models support widespread post-transcriptional regulation of protein abundance by interacting partners 94%
"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.