Ultrafast and accurate 16S rRNA microbial community analysis using Kraken 2

Abstract

Background: For decades, 16S ribosomal RNA sequencing has been the primary means for identifying the bacterial species present in a sample with unknown composition. One of the most widely used tools for this purpose today is the QIIME (Quantitative Insights Into Microbial Ecology) package. Recent results have shown that the newest release, QIIME 2, has higher accuracy than QIIME, MAPseq, and mothur when classifying bacterial genera from simulated human gut, ocean, and soil metagenomes, although QIIME 2 also proved to be the most computationally expensive. Kraken, first released in 2014, has been shown to provide exceptionally fast and accurate classification for shotgun metagenomics sequencing projects. Bracken, released in 2016, then provided users with the ability to accurately estimate species or genus relative abundances using Kraken classification results. Kraken 2, which matches the accuracy and speed of Kraken 1, now supports 16S rRNA databases, allowing for direct comparisons to QIIME and similar systems. Methods: For a comprehensive assessment of each tool, we compare the computational resources and speed of QIIME 2's q2-feature-classifier, Kraken 2, and Bracken in generating the three main 16S rRNA databases: Greengenes, SILVA, and RDP. For an evaluation of accuracy, we evaluated each tool using the same simulated 16S rRNA reads from human gut, ocean, and soil metagenomes that were previously used to compare QIIME, MAPseq, mothur, and QIIME 2. We evaluated accuracy based on the accuracy of the final genera read counts assigned by each tool. Finally, as Kraken 2 is the only tool providing per-read taxonomic assignments, we evaluate the sensitivity and precision of Kraken 2's per-read classifications. Results: For both the Greengenes and SILVA database, Kraken 2 and Bracken are up to 100 times faster at database generation. For classification, using the same data as previous studies, Kraken 2 and Bracken are up to 300 times faster, use 100x less RAM, and generate results that more accurate at 16S rRNA profiling than QIIME 2's q2-feature-classifier. Conclusion: Kraken 2 and Bracken provide a very fast, efficient, and accurate solution for 16S rRNA metataxonomic data analysis.

Measuring the microbiome: Best practices for developing and benchmarking microbiomics methods

“…In some cases, a ground truth may not be necessary, for example to measure the consistency among methods rather than their accuracy [170] Test data should be appropriate to the intended usage, and for the different use cases that a user can reasonably expect to encounter (if those different use cases can be expected to impact performance). This may include sample type, degrees of microbial diversity/complexity, or different targets (e.g., marker genes) [67] , [78] , [147] , [171] . Multiple datasets and types of data should be used for method testing and optimization.…”

“…Method users should be aware that testing on a single dataset could lead to overfitting (as discussed above), and such an “optimized” method may not generalize well to other data. Benchmarks that test on multiple datasets and types of data demonstrate the variation in performance across multiple systems and technologies [67] , [147] , [171] . Selecting methods for comparison to a new method or in a benchmarking study should be performed judiciously.…”

“…Typically the methods used in this comparison should represent one or more “gold standards”, i.e. methods that are both popular in the field and demonstrated to have superior accuracy according to previous benchmarks [113] , [171] . Performance metrics should be selected to evaluate criteria that are important to users, e.g.…”

“…Test data should be appropriate to the intended usage, and for the different use cases that a user can reasonably expect to encounter (if those different use cases can be expected to impact performance). This may include sample type, degrees of microbial diversity/complexity, or different targets (e.g., marker genes) [67] , [78] , [147] , [171] .…”

Circular Economy of Anaerobic Biofilm Microbiomes: A Meta-Analysis Framework for Re-exploration of Amplicon Sequencing Data

“…Several processing pipelines have been developed with a series of steps and programs employed to align, denoise, and remove spurious sequences (i.e. MOTHUR [28], Qiime [29], KRAKEN [30]. While these pipelines are designed and updated for quality assurance, users typically follow a series of steps without deviation.…”