KnvResGAT: SARS-CoV-2 sequence classification using k-mer natural vector and graph attention networks.

Area of Science:

• Genomics
• Bioinformatics
• Machine Learning

Background:

• Accurate and efficient classification of SARS-CoV-2 lineages is crucial for epidemiological surveillance and public health.
• Existing methods may face challenges in generalization and scalability with the increasing volume of genomic data.

Purpose of the Study:

• To develop an efficient and accurate computational method for SARS-CoV-2 lineage classification.
• To leverage k-mer Natural Vector (KNV) representations and Graph Attention Networks (GAT) for improved classification performance.

Main Methods:

• Proposing KnvResGAT, which combines k-mer Natural Vector (KNV) representations with a residual multi-head Graph Attention Network (GAT).
• Constructing a k-nearest-neighbor (kNN) similarity graph in the KNV feature space for classification.
• Evaluating the method on a large dataset of 182,851 SARS-CoV-2 genomes across 103 Pango lineages using a time-aware split.

Main Results:

• KnvResGAT achieved an accuracy of 0.9729 and a Macro-F1 score of 0.9636 on the curated dataset.
• The proposed method outperformed established tools like Pangolin (0.9673 accuracy, 0.9471 Macro-F1) and a ResMLP baseline (0.9654 accuracy, 0.9520 Macro-F1).
• Demonstrated improved generalization capabilities for multi-class SARS-CoV-2 lineage classification.

Conclusions:

• KnvResGAT offers a highly accurate and efficient approach for SARS-CoV-2 lineage classification.
• The combination of KNV and GAT provides a powerful framework for analyzing viral genomic data.
• This method has the potential to enhance real-time epidemiological surveillance and response efforts.