Reads2Vec: Efficient Embedding of Raw High-Throughput Sequencing Reads Data.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• The COVID-19 pandemic generated massive SARS-CoV-2 genomic data, overwhelming traditional analysis methods.
• Existing tools like Pangolin require assembled, aligned sequences, which can be a bottleneck with advancing sequencing technologies.

Purpose of the Study:

• To develop a novel alignment-free method for direct analysis of raw SARS-CoV-2 sequencing reads.
• To create a numerical representation (embedding) of genomic data for improved classification and clustering.

Main Methods:

• Proposed Reads2Vec, an alignment-free embedding approach generating fixed-length feature vectors from raw sequencing reads.
• Applied embedding to classification and clustering algorithms.
• Evaluated performance on simulated and real SARS-CoV-2 genomic data.

Main Results:

• Reads2Vec demonstrated superior classification and clustering performance on simulated data compared to alignment-free baselines.
• Alignment-free embeddings showed better clustering properties than the Pangolin tool on real data.
• The spike region of the SARS-CoV-2 genome was identified as a key driver of alignment-free clustering.

Conclusions:

• Reads2Vec offers an efficient alternative for analyzing large-scale SARS-CoV-2 genomic data directly from raw reads.
• Alignment-free methods, particularly when focusing on key genomic regions like the spike, provide valuable insights for viral dynamics and classification.