COVID-DeepPredictor: Recurrent Neural Network to Predict SARS-CoV-2 and Other Pathogenic Viruses

Area of Science:

• Bioinformatics
• Genomic sequence analysis
• Machine learning in virology

Background:

• The COVID-19 pandemic, caused by SARS-CoV-2, highlights the need for rapid and accurate viral identification.
• The polymorphic nature of SARS-CoV-2 and the presence of other pathogens (e.g., SARS-CoV-1, MERS-CoV, Ebola, Dengue, Influenza) necessitate reliable genomic predictors.
• Distinguishing between various viral pathogens is crucial for effective treatment and public health response.

Purpose of the Study:

• To propose COVID-DeepPredictor, a deep learning model for identifying unknown pathogen sequences.
• To leverage Long Short Term Memory (LSTM) recurrent neural networks for sequence prediction.
• To develop an alignment-free technique for efficient viral sequence analysis.

Main Methods:

• Utilized the k-mer technique to generate Bag-of-Descriptors (BoDs) and Bag-of-Unique-Descriptors (BoUDs) for sequence representation.
• Employed Long Short Term Memory (LSTM) networks within a deep learning framework.
• Validated the model using k-fold cross-validation and tested on unseen datasets of SARS-CoV-2 and other viral sequences.

Main Results:

• Achieved 100% prediction accuracy on the validation dataset.
• Demonstrated high accuracy on test datasets, ranging from 99.51% to 99.94%.
• Outperformed existing state-of-the-art methods including Linear Discriminant Analysis, Random Forests, and Gradient Boosting.

Conclusions:

• COVID-DeepPredictor offers a highly accurate and efficient method for identifying viral pathogens based on genomic sequences.
• The alignment-free, deep learning approach provides a robust solution for distinguishing SARS-CoV-2 from other viruses.
• The study provides valuable insights into optimizing k-mer parameters and demonstrates superior performance compared to traditional methods and Nucleotide BLAST.