MP-VHPPI: Meta predictor for viral host protein-protein interaction prediction in multiple hosts and viruses

Muhammad Nabeel Asim, Ahtisham Fazeel, Muhammad Ali Ibrahim, Andreas Dengel, Sheraz AhmedFrontiers in Medicine 2022 · DOI

TL;DR

This paper introduces MP-VHPPI, a novel meta-predictor designed to accurately predict interactions between viral and host proteins across a wide range of viruses and hosts. By combining advanced sequence encoding techniques with a multi-stage machine learning approach, MP-VHPPI significantly outperforms existing predictors, offering a more robust tool for understanding viral pathogenesis and developing antiviral strategies.

What Problem Does This Paper Address?

Understanding how viruses interact with host proteins (Viral-Host Protein-Protein Interactions, or VHPPIs) is crucial for deciphering viral mechanisms, host immune responses, and for developing effective treatments and vaccines. While many AI-based predictors exist, they often perform well only for specific virus-host pairs. This limitation hinders their generalizability and broad applicability in tackling diverse viral diseases. The challenge is to create a single predictor that can accurately identify VHPPIs across a wide spectrum of viruses and hosts, overcoming the specificity limitations of current methods.

Methods

The MP-VHPPI predictor employs a multi-stage approach:

  • Feature Representation: It utilizes two sophisticated sequence encoding methods:
  • Amphiphilic Pseudo-Amino Acid Composition (APAAC): Captures hydrophobicity, hydrophilicity, and side chain mass to represent protein sequences.
  • Quasi-sequence (QS) Order: Leverages physicochemical properties like hydrophobicity, hydrophilicity, polarity, and side chain volume to encode sequence order and distributional information. Crucially, the authors don't use all properties from these encoders but employ a "forward feature selection" strategy to identify the most informative physicochemical properties for each encoder.

  • Dimensionality Reduction: To handle potential redundancy and extract more discriminative features, a feature agglomeration technique is used to reduce the dimensionality of the generated feature space.

  • Iterative Representation Learning (Two-Stage Prediction):
    1. Stage 1: Statistical vectors from APAAC and QS order (individually and combined, after dimensionality reduction) are fed into Random Forest (RF) and Extra Trees (ET) classifiers. The positive class probabilities generated by these classifiers form a new feature space.
  • Stage 2: This new probabilistic feature space is then used to train a Support Vector Machine (SVM) classifier, which makes the final VHPPI predictions. This iterative approach allows the model to learn more robust and discriminative representations.

Datasets

The study utilizes 7 publicly available benchmark datasets that have been previously used to evaluate VHPPI predictors. These datasets are derived from studies by Barman et al. (22), Fatma et al. (29), Yang et al. (36), and Zhou et al. (30). The datasets cover a range of viruses and hosts, including human proteins, SARS-CoV-2, Ebola virus, H1N1 virus, and many others, with some datasets specifically designed for evaluating generalization to unseen viruses.

Key Insights and Ideas

  • Property Selection is Key: Not all physicochemical properties within APAAC and QS order are equally informative. Identifying and selecting the most relevant properties significantly impacts prediction performance.
  • Dimensionality Reduction Enhances Features: Feature agglomeration helps to remove redundant information, leading to a more informative feature space and improved classifier performance.
  • Iterative Learning Boosts Accuracy: The two-stage prediction process, where the outputs of RF/ET classifiers are used as input for SVM, is critical for achieving higher accuracy and robustness. This "meta-predictor" approach leverages the strengths of multiple models.
  • Generalizability is Achievable: By employing robust feature engineering and a meta-prediction strategy, the MP-VHPPI demonstrates improved performance across diverse datasets, including those with unseen viruses, addressing a major limitation of prior work.

Key Results

The MP-VHPPI predictor demonstrated superior performance across 7 benchmark datasets. On average, it outperformed existing VHPPI predictors by:

  • 3.07% in Accuracy
  • 6.07% in Mathews Correlation Coefficient (MCC)
  • 2.95% in Precision
  • 2.85% in Sensitivity

The paper highlights significant performance gains on datasets like SARS-CoV-2 and those designed to test generalization to unseen viruses, where MP-VHPPI showed substantial improvements over existing methods.

Limitations and Future Work

  • Bias Towards Type II Error: While MP-VHPPI shows good overall performance, the authors acknowledge that it is still biased towards type II errors (false negatives).
  • Future Work: The authors suggest optimizing the predictive pipeline further to enhance robustness and reduce this bias. They also propose exploring representations learned through advanced language models like BERT and XLNET to potentially further improve performance.

A web server for the MP-VHPPI predictor is available at https://sds_genetic_analysis.opendfki.de/MP-VHPPI/ to facilitate its use by the scientific community.

Summary updated

Abstract

Viral-host protein-protein interaction (VHPPI) prediction is essential to decoding molecular mechanisms of viral pathogens and host immunity processes that eventually help to control the propagation of viral diseases and to design optimized therapeutics. Multiple AI-based predictors have been developed to predict diverse VHPPIs across a wide range of viruses and hosts, however, these predictors produce better performance only for specific types of hosts and viruses. The prime objective of this research is to develop a robust meta predictor (MP-VHPPI) capable of more accurately predicting VHPPI across multiple hosts and viruses. The proposed meta predictor makes use of two well-known encoding methods Amphiphilic Pseudo-Amino Acid Composition (APAAC) and Quasi-sequence (QS) Order that capture amino acids sequence order and distributional information to most effectively generate the numerical representation of complete viral-host raw protein sequences. Feature agglomeration method is utilized to transform the original feature space into a more informative feature space. Random forest (RF) and Extra tree (ET) classifiers are trained on optimized feature space of both APAAC and QS order separate encoders and by combining both encodings. Further predictions of both classifiers are utilized to feed the Support Vector Machine (SVM) classifier that makes final predictions. The proposed meta predictor is evaluated over 7 different benchmark datasets, where it outperforms existing VHPPI predictors with an average performance of 3.07, 6.07, 2.95, and 2.85% in terms of accuracy, Mathews correlation coefficient, precision, and sensitivity, respectively. To facilitate the scientific community, the MP-VHPPI web server is available at https://sds_genetic_analysis.opendfki.de/MP-VHPPI/.
BibTeX
@inproceedings{asim2022mp,
  author = {Muhammad Nabeel Asim and Ahtisham Fazeel and Muhammad Ali Ibrahim and Andreas Dengel and Sheraz Ahmed},
  title = {{MP-VHPPI: Meta predictor for viral host protein-protein interaction prediction in multiple hosts and viruses}},
  booktitle = {Frontiers in Medicine},
  year = {2022},
  doi = {10.3389/fmed.2022.1025887},
}
Figure