Efficient and Accurate OTU Clustering with GPU-Based Sequence Alignment and Dynamic Dendrogram Cutting.

Area of Science:

• Microbiology
• Bioinformatics
• Computational Biology

Background:

• De novo clustering is essential for microbial taxonomic profiling using 16S rRNA amplicon sequencing.
• Existing operational taxonomic unit (OTU) clustering pipelines often rely on a fixed 97% sequence similarity cutoff, which can limit accuracy.
• Large datasets pose computational challenges for traditional dendrogram-based clustering methods.

Purpose of the Study:

• To introduce CRiSPy, a novel dendrogram-based pipeline for de novo OTU clustering.
• To improve the accuracy of taxonomic profiling by employing a dynamic distance cutoff.
• To enhance computational efficiency for processing large microbial datasets.

Main Methods:

• CRiSPy utilizes an anomaly detection technique on dendrogram merging heights to determine a dynamic distance cutoff.
• It employs hierarchical clustering on a genetic distance matrix derived from all-against-all read comparisons.
• Efficient algorithms, including GPU-accelerated parallel processing and memory-efficient hierarchical clustering, were developed to handle large datasets.

Main Results:

• CRiSPy achieves more accurate OTU groupings compared to existing OTU clustering applications.
• The pipeline effectively processes datasets larger than 10,000 unique reads, overcoming computational limitations.
• Anomaly detection provides a more robust method for setting distance cutoffs in OTU clustering.

Conclusions:

• CRiSPy offers a significant advancement in de novo OTU clustering for microbial community analysis.
• The dynamic distance cutoff approach improves taxonomic profiling accuracy.
• CRiSPy's efficient algorithms make it suitable for analyzing large-scale microbial sequencing data.