Identifying New COVID-19 Variants from Spike Proteins Using Novelty Detection

Area of Science:

• Virology
• Computational Biology
• Machine Learning

Background:

• The COVID-19 pandemic continues with the emergence of new viral variants.
• Machine learning has proven valuable in analyzing the virus and its impact.
• Early detection of variants is crucial for public health responses.

Purpose of the Study:

• To apply novelty detection algorithms for identifying specific COVID-19 variants.
• To evaluate the accuracy of these algorithms on viral spike protein sequences.

Main Methods:

• Utilized novelty detection algorithms, specifically One Class SVM.
• Applied fine-tuned parameters to analyze ProtVec unaligned COVID-19 spike protein sequences.
• Tested algorithms on B.1.1.7 and B.1.351 variants.

Main Results:

• Achieved 79.64% accuracy for the B.1.1.7 variant.
• Achieved 82.43% accuracy for the B.1.351 variant.
• Demonstrated the effectiveness of machine learning in variant detection.

Conclusions:

• Automated and timely detection of COVID-19 variants is feasible using machine learning.
• This technology can support the development of targeted medicines and vaccines.
• Facilitates informed public health strategies and pandemic mitigation efforts.