ClassGraph: Improving Metagenomic Read Classification with Overlap Graphs.

Area of Science:

• Microbiology
• Bioinformatics
• Computational Biology

Background:

• Environmental sequencing generates vast amounts of microbial data.
• Accurate taxonomic annotation of sequencing reads is crucial for understanding microbial communities.
• Current k-mer based methods achieve high precision but suffer from low sensitivity, particularly with divergent genomes.

Purpose of the Study:

• To develop a novel computational method, ClassGraph, for improving taxonomic classification of microbial sequencing reads.
• To enhance the sensitivity of read classification while maintaining high precision.
• To address limitations of existing methods in classifying divergent or mutated genomes, such as those from viruses.

Main Methods:

• ClassGraph utilizes a read overlap graph to represent sequence relationships.
• A label propagation algorithm is applied to the graph to refine taxonomic assignments.
• The method integrates and refines results from existing taxonomic classification tools.

Main Results:

• ClassGraph demonstrated improved sensitivity and F-measure across simulated and real microbial datasets.
• The method maintained high precision, comparable to existing tools.
• Significant improvements in classification accuracy were observed for challenging datasets, including viral and complex environmental samples, where traditional tools classified less than 40% of reads.

Conclusions:

• ClassGraph effectively enhances taxonomic classification accuracy and sensitivity in microbial community analysis.
• The approach is particularly beneficial for classifying reads from highly mutated or divergent genomes.
• ClassGraph offers a valuable tool for improving the analysis of complex environmental metagenomic data.