Clustering Metagenomic Sequences with Interpolated Markov Models
Author Information
Author(s): Kelley David R, Salzberg Steven L
Primary Institution: University of Maryland
Hypothesis
Can unsupervised sequence clustering methods improve the accuracy of metagenomic data analysis?
Conclusion
SCIMM and PHYSCIMM are highly accurate methods for clustering metagenomic sequences, with SCIMM being ideal for novel microbes and PHYSCIMM improving accuracy when related training genomes are available.
Supporting Evidence
- SCIMM outperformed previous unsupervised methods in clustering accuracy.
- PHYSCIMM showed improved performance when related training genomes were available.
- Clustering accuracy was measured using precision, recall, and adjusted Rand index.
- Longer read lengths improved clustering accuracy.
- SCIMM is effective for environments with ten or fewer species.
Takeaway
This study introduces two methods, SCIMM and PHYSCIMM, that help group similar DNA sequences from environmental samples, making it easier to understand the diversity of microbes.
Methodology
The study uses interpolated Markov models (IMMs) for unsupervised clustering of DNA sequences, comparing the performance of SCIMM and a hybrid method called PHYSCIMM.
Potential Biases
The reliance on public databases for training genomes may introduce bias if those genomes are not representative of the metagenomic samples.
Limitations
The performance of unsupervised clustering methods degrades as the number of genomes increases, and SCIMM may struggle with low abundance species.
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website