Background: Illumina sequencing of a marker gene is popular in metagenomic studies. However, Illumina paired-end (PE) reads sometimes cannot be merged into single reads for subsequent analysis. When mergeable PE reads are limited, one can simply use only first reads for taxonomy annotation, but that wastes information in the second reads. Presumably, including second reads should improve taxonomy annotation. However, a rigorous investigation of how best to do this and how much can be gained has not been reported. Results: We evaluated two methods of joining as opposed to merging PE reads into single reads for taxonomy annotation using simulated data with sequencing errors. Our rigorous evaluation involved several top classifiers (RDP classifier, SINTAX, and two alignment-based methods) and realistic benchmark datasets. For most classifiers, read joining ameliorated the impact of sequencing errors and improved the accuracy of taxonomy predictions. For alignment-based top-hit classifiers, rearranging the reference sequences is recommended to avoid improper alignments of joined reads. For word-counting classifiers, joined reads could be compared to the original reference for classification. We also applied read joining to our own real MiSeq PE data of nasal microbiota of asthmatic children. Before joining, trimming low quality bases was necessary for optimizing taxonomy annotation and sequence clustering. We then showed that read joining increased the amount of effective data for taxonomy annotation. Using these joined trimmed reads, we were able to identify two promising bacterial genera that might be associated with asthma exacerbation. Conclusions: When mergeable PE reads are limited, joining them into single reads for taxonomy annotation is always recommended. Reference sequences may need to be rearranged accordingly depending on the classifier. Read joining also relaxes the constraint on primer selection, and thus may unleash the full capacity of Illumina PE data for taxonomy annotation. Our work provides guidance for fully utilizing PE data of a marker gene when mergeable reads are limited.
All Science Journal Classification (ASJC) codes