Disentangling cobionts and contamination in long-read genomic data using sequence composition | JoVE Visualize

Area of Science:

Genomics
Bioinformatics
Computational Biology

Background:

Genome sequencing projects are expanding to diverse life forms, creating computational challenges in analyzing mixed-species samples.
Environmental samples frequently contain target organisms alongside symbiotic partners (cobionts) and contaminants, necessitating effective sequence separation methods.
Current reference-based methods struggle with underrepresented eukaryotic taxa, highlighting the need for alternative approaches.

Purpose of the Study:

To develop and evaluate a computational strategy for distinguishing between target organisms, cobionts, and contaminants in sequencing data.
To minimize dependence on comprehensive reference databases by utilizing intrinsic sequence composition differences.
To enable the analysis of complex environmental samples, including those from the Darwin Tree of Life project.

Main Methods:

Utilizing variational autoencoders to learn two-dimensional representations of read tetranucleotide composition.
Visualizing these learned embeddings to identify distinct organismal components within a sample.
Integrating additional data like coding density, coverage, and taxonomic labels for annotation and assessment.
Applying the method to large-scale insect genome data from the Darwin Tree of Life project.

Main Results:

Distinct clusters representing different organisms were successfully visualized in the sequence composition embeddings.
The approach demonstrated scalability to millions of sequences, enabling analysis of unassembled read sets.
Interactive visualization facilitated rapid assessment and identification of sample components.
A significant portion of cobionts identified by reference-based screening were corroborated, and novel genomes were retrieved.

Conclusions:

Sequence composition analysis, particularly using dimensionality reduction techniques like variational autoencoders, offers a powerful, reference-light approach for dissecting complex environmental sequencing data.
This method effectively identifies symbionts and contaminants and aids in retrieving genomes lacking robust reference data.
The approach is scalable and valuable for large-scale genome sequencing initiatives like the Darwin Tree of Life project.