Mutation prediction in the SARS-CoV-2 genome using attention-based neural machine translation

Area of Science:

• Virology
• Computational Biology
• Genomics

Background:

• Severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) exhibits rapid evolution and frequent mutations, complicating containment efforts.
• Viral genomic changes can lead to increased resistance against current vaccines and antiviral therapies.
• Proactive prediction of viral mutations is crucial for developing effective countermeasures and mitigating pandemic impact.

Purpose of the Study:

• To propose and evaluate novel neural machine translation (NMT) architectures for predicting mutations in SARS-CoV-2.
• To investigate the application of recurrent neural networks (RNNs) for forecasting viral evolution in specific SARS-CoV-2 non-structural proteins (NSPs).

Main Methods:

• Development of multiple NMT architectures utilizing recurrent neural networks (RNNs).
• Creation and pre-processing of sequence pairs using k-means clustering and nearest neighbors for NMT model training.
• Exploration of NMT training methodologies for extended biological sequences.
• Evaluation and benchmarking of the proposed NMT models for efficiency and reliability.

Main Results:

• Demonstration of the efficiency and reliability of RNN-based NMT models in predicting SARS-CoV-2 mutations.
• Successful application of NMT techniques to analyze mutations in selected SARS-CoV-2 non-structural proteins (NSP1, NSP3, NSP5, NSP8, NSP9, NSP13, NSP15).

Conclusions:

• The proposed NMT approaches show promise for accurately predicting viral mutations.
• This predictive capability can significantly enhance preparedness against emerging and potentially more virulent SARS-CoV-2 variants.
• The study provides valuable insights into applying NMT for analyzing long biological sequences and understanding viral evolution.