Variant-driven early warning via unsupervised machine learning analysis of spike protein mutations for COVID-19

Area of Science:

• Virology and Bioinformatics
• Epidemiology
• Machine Learning

Background:

• Unprecedented genomic data availability for COVID-19 allows real-time tracking of viral evolution.
• The Spike protein is crucial for viral entry and replication, making it a key target for studying mutations.
• Understanding variant emergence is critical for managing the multi-wave nature of the pandemic.

Purpose of the Study:

• To develop a machine learning algorithm for real-time identification and characterization of SARS-CoV-2 variants.
• To establish a framework for predicting the emergence of new variants of epidemiological interest.
• To correlate variant evolution with the temporal patterns of COVID-19 pandemic waves.

Main Methods:

• Utilized Levenshtein distance on Spike protein sequences to cluster viral data.
• Developed a machine learning algorithm for temporal clustering and variant identification.
• Defined persistent variants as stable sequence chains and emerging variants as branching events.

Main Results:

• Successfully identified and defined persistent and emerging variants consistent with known evidence.
• The algorithm provides an early warning when a variant cluster reaches 1% of sequence data.
• Validated the approach on the Alpha variant and predicted AY.4.2 ('Delta plus') as an emerging variant.

Conclusions:

• The developed algorithm effectively tracks viral evolution and identifies variants of concern.
• A strong correlation exists between the emergence of new variants and the multi-wave pattern of the pandemic.
• The Mutation Epidemiological Renormalisation Group framework aids in understanding variant epidemiology.