Machine Learning Algorithms Associate Case Numbers with SARS-CoV-2 Variants Rather Than with Impactful Mutations.

Area of Science:

• Epidemiology
• Genomics
• Machine Learning

Background:

• COVID-19 pandemic spurred efforts to predict case numbers using epidemiological data.
• Viral genomic information, including variant-specific virulence, was largely overlooked in predictive models.
• The Alpha and Delta variants of SARS-CoV-2 co-circulated, offering an opportunity to study genomic impacts on case prediction.

Purpose of the Study:

• To investigate whether viral genomic sequences can improve predictions of future SARS-CoV-2 case numbers.
• To assess the predictive performance of machine learning models trained on genomic data.
• To analyze the explainability of these models to understand their predictive basis.

Main Methods:

• Implemented simple predictive models using genomic sequences of SARS-CoV-2 Alpha and Delta variants.
• Encoded viral sequences and matched them with future case numbers based on collection dates.
• Trained two algorithms: random forests and a feed-forward neural network.

Main Results:

• Models achieved high prediction accuracies (≥93%).
• Explainability analysis revealed models associated predictions with specific variants (Alpha, Delta), not virulence-linked mutations.
• The models did not identify specific mutations correlating with increased case numbers.

Conclusions:

• While genomic data can predict case numbers, models may rely on variant identity rather than specific virulence factors.
• Explainability analysis is crucial to ensure predictive models are not misleading.
• Further research is needed to understand the interplay between viral genomics and epidemiological predictions.