The optimal metric for viral genome space

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Genome structural similarity is crucial for classification and phylogenetic analysis.
• Alignment-free methods are increasingly important due to the growing number of available genomes.
• The natural vector method represents sequences as vectors using statistical moments for clustering.

Purpose of the Study:

• To address the challenge of determining optimal metrics for combining elements in natural vectors.
• To develop a rigorous theoretical framework for weighting k-mers and orders in natural vectors.
• To enhance the accuracy of genome classification using alignment-free approaches.

Main Methods:

• Transforming the determination of optimal weights into an optimization problem.
• Resolving the optimization problem through gradient-based techniques.
• Applying the optimized natural vector method to genome classification tasks.

Main Results:

• Achieved a substantial improvement in classification accuracy, reaching 92.73% on the testing set.
• Outperformed other existing alignment-free methods in classification tasks.
• Demonstrated the effectiveness of optimal weights in enhancing the natural vector method.

Conclusions:

• The developed method provides an outstanding metric for virus classification.
• Offers valuable insights into feature integration strategies for alignment-free bioinformatics methods.
• Highlights the potential of optimization techniques for improving genomic analysis tools.