ViralVectors: compact and scalable alignment-free virome feature generation

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• The exponential growth of viral sequencing data, particularly for SARS-CoV-2, necessitates advanced computational methods for genomic surveillance.
• Existing methods struggle with the heterogeneous nature of sequencing data, including raw, unaligned, or unassembled reads.

Purpose of the Study:

• To introduce ViralVectors, a novel method for generating compact feature vectors from diverse virome sequencing data.
• To enable effective and timely downstream analyses, such as classification and clustering, for large-scale viral genomic surveillance.

Main Methods:

• Utilizing minimizers, a lightweight sequence signature technique, for compact feature vector generation from raw or processed sequencing reads.
• Applying ViralVectors to heterogeneous datasets, including large-scale SARS-CoV-2 spike sequences, Coronaviridae spike sequences, and raw whole-genome sequencing reads from nasal swabs.

Main Results:

• ViralVectors demonstrates scalability with 2.5 million SARS-CoV-2 spike sequences.
• The method shows robustness with 3,000 Coronaviridae spike sequences and effectiveness in processing 4,000 raw whole-genome sequencing read sets.
• ViralVectors outperforms existing benchmarks in classification and clustering tasks on various sequencing data types.

Conclusions:

• ViralVectors provides an efficient and scalable solution for analyzing large and diverse viral sequencing datasets.
• The minimizer-based approach offers a powerful new tool for genomic surveillance and understanding viral evolution.
• This method facilitates timely decision-making in public health by enabling rapid analysis of viral genomic data.