Yesterday, Molecular Ecology Resources put online an unedited version of a recent paper which I co-authored. This time, Rodney Richardson at Ohio State University has made a tremendous work of evaluating three taxonomic classification software – the RDP Naïve Bayesian Classifier, RTAX and UTAX – on a set of DNA barcoding regions commonly used for plants, namely the ITS2, and the matK, rbcL, trnL and trnH genes.
In the paper (1), we discuss the results, merits and limitations of the classifiers. In brief, we found that:
- There is a considerable trade-off between accuracy and sensitivity for the classifiers tested, which indicates a need for improved sequence classification tools (2)
- UTAX was superior with respect to error rate, but was exceedingly stringent and thus suffered from a low assignment rate
- The RDP Naïve Bayesian Classifier displayed high sensitivity and low error at the family and order levels, but had a genus-level error rate of 9.6 percent
- RTAX showed high sensitivity at all taxonomic ranks, but at the same time consistently produced the high error rates
- The choice of locus has significant effects on the classification sensitivity of all tested tools
- All classifiers showed strong relationships between database completeness, classification sensitivity and classification accuracy
We believe that the methods of comparison we have used are simple and robust, and thereby provides a methodological and conceptual foundation for future software evaluations. On a personal note, I will thoroughly enjoy working with Rodney and Reed again; I had a great time discussing the ins and outs of taxonomic classification with them! The paper can be found here.
References and notes
- Richardson RT, Bengtsson-Palme J, Johnson RM: Evaluating and Optimizing the Performance of Software Commonly Used for the Taxonomic Classification of DNA Sequence Data. Molecular Ecology Resources, Early view (2016). doi: 10.1111/1755-0998.12628 [Paper link]
- This is something that several classifiers also showed in the evaluation we did for the Metaxa2 paper (3). Interestingly enough, Metaxa2 is better at maintaining high accuracy also as sensitivity is increased.
- Bengtsson-Palme J, Hartmann M, Eriksson KM, Pal C, Thorell K, Larsson DGJ, Nilsson RH: Metaxa2: Improved identification and taxonomic classification of small and large subunit rRNA in metagenomic data. Molecular Ecology Resources, 15, 6, 1403–1414 (2015). doi: 10.1111/1755-0998.12399 [Paper link]
After almost a year in different stages of review and revision, in which the paper (but not the software) saw a total transformation, I am happy to announce that the paper describing Metaxa2 has been accepted in Molecular Ecology Resources and is available in a rudimentary online early form. The figures in this version are not that pretty, but those who wants to read the paper asap, you have the possibility to do so.
This means that if you have been using Metaxa2 for a publication, there is now a new preferred way of citing this, namely:
Bengtsson-Palme J, Hartmann M, Eriksson KM, Pal C, Thorell K, Larsson DGJ, Nilsson RH: Metaxa2: Improved Identification and Taxonomic Classification of Small and Large Subunit rRNA in Metagenomic Data. Molecular Ecology Resources (2015). doi: 10.1111/1755-0998.12399
The paper (1), apart from describing the new Metaxa version, also brings a very thorough evaluation of the software, compared to other tools for taxonomic classification implemented in QIIME (2). In short, we show that:
- Metaxa2 can make trustworthy taxonomic classifications even with reads as short as 100 bp
- Generally, the performance is reliable across the entire SSU rRNA gene, regardless of which V-region a read is derived from
- Metaxa2 can reliably recapture species composition from short-read metagenomic data, comparable with results of amplicon sequencing
- Metaxa2 outperforms other popular tools such as Mothur (3), the RDP Classifier (4), Rtax (5) and the QIIME implementation of Uclust (6) in terms of proportion of correctly classified reads from metagenomic data
- The false positive rate of Metaxa2 is very close to zero; far superior to many of the above mentioned tools, many of which assume that reads must derive from the rRNA gene
Metaxa2 can be downloaded here. We have already used it for around two years internally, and it forms the base of the taxonomic classifications in e.g. our recently published paper on antibiotic resistance in a polluted Indian lake (7).
- Bengtsson-Palme J, Hartmann M, Eriksson KM, Pal C, Thorell K, Larsson DGJ, Nilsson RH: Metaxa2: Improved Identification and Taxonomic Classification of Small and Large Subunit rRNA in Metagenomic Data. Molecular Ecology Resources (2015). doi: 10.1111/1755-0998.12399 [Paper link]
- Caporaso JG, Kuczynski J, Stombaugh J et al.: QIIME allows analysis of high-throughput community sequencing data. Nature Methods, 7, 335–336 (2010).
- Schloss PD, Westcott SL, Ryabin T et al.: Introducing mothur: open-source, platform-independent, community-supported software for describing and comparing microbial communities. Applied and Environmental Microbiology, 75, 7537–7541 (2009).
- Wang Q, Garrity GM, Tiedje JM, Cole JR: Naive Bayesian classifier for rapid assignment of rRNA sequences into the new bacterial taxonomy. Applied and Environmental Microbiology, 73, 5261–5267 (2007).
- Soergel DAW, Dey N, Knight R, Brenner SE: Selection of primers for optimal taxonomic classification of environmental 16S rRNA gene sequences. The ISME Journal, 6, 1440–1444 (2012).
- Edgar RC: Search and clustering orders of magnitude faster than BLAST. Bioinformatics, 26, 2460–2461 (2010).
- Bengtsson-Palme J, Boulund F, Fick J, Kristiansson E, Larsson DGJ: Shotgun metagenomics reveals a wide array of antibiotic resistance genes and mobile elements in a polluted lake in India. Frontiers in Microbiology, 5, 648 (2014).