Assessing the Resilience of Machine Learning Classification Algorithms on SARS-CoV-2 Genome Sequences Generated with...

Area of Science:

• Genomics
• Bioinformatics
• Machine Learning

Background:

• Third-generation sequencing (TGS) generates long reads crucial for genome assembly, including SARS-CoV-2.
• High error rates in TGS data can compromise genome assembly and biological interpretation.
• Understanding SARS-CoV-2 evolution and transmission relies on accurate genome sequencing.

Purpose of the Study:

• To evaluate machine learning (ML) models for analyzing SARS-CoV-2 genome sequences with sequencing errors.
• To compare the effectiveness of six different embedding techniques for error-incorporated genome data.
• To identify robust ML approaches for accurate SARS-CoV-2 genome sequence identification.

Main Methods:

• Utilized ML models with six distinct embedding techniques.
• Analyzed SARS-CoV-2 genome sequences with simulated and random errors mimicking TGS profiles.
• Employed spaced k-mers and weighted k-mers embedding methods for vector generation.

Main Results:

• Spaced k-mers embedding demonstrated high accuracy in classifying error-free SARS-CoV-2 genomes.
• Spaced k-mers and weighted k-mers embedding methods showed high accuracy in predicting error-incorporated sequences.
• Fixed-length vectors generated by these methods contributed to high model performance.

Conclusions:

• The study highlights the efficacy of specific ML embedding techniques for handling TGS errors in SARS-CoV-2 genomes.
• Researchers can leverage these findings to select appropriate ML models for accurate genomic analysis.
• This work provides insights into improving the identification of critical SARS-CoV-2 genome sequences.