Algorithm for post-clustering curation of DNA amplicon data yields reliable biodiversity estimates.

Frøslev TG, Kjøller R, Bruun HH, Ejrnæs R, Brunbjerg AK, Pietroni C, Hansen AJ. · 2017

Read Algorithm for post-clustering curation of DNA amplicon data yields reliable biodiversity estimates. by Frøslev TG, Kjøller R, Bruun HH, Ejrnæs R, Brunbjerg AK, Pietroni C, Hansen AJ. (2017) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/europmc-pmc5662604.txt.

Introduction

Frøslev and co-authors present a tool that cleans up DNA data used to count species. Their method, called LULU, works on data from DNA metabarcoding. Such data group DNA reads into units called OTUs. Errors from lab steps create false units and inflate species counts.

LULU spots false units by mixing two clues. It checks how alike two units are and whether they occur together across samples. The team tested it on soil DNA from 130 sites in Denmark. Plant survey data from the field served as the check. LULU improved diversity estimates across several counting methods. It needs no reference database to work.

Catalog record

CREATORFrøslev TG, Kjøller R, Bruun HH, Ejrnæs R, Brunbjerg AK, Pietroni C, Hansen AJ.
YEAR2017
COLLECTIONPAPERS
FORMATTXT
FILE SIZE78 KB
LICENSECC BY 4.0
RIGHTSCC BY
SOURCEEurope PMC

Source: https://europepmc.org/articles/PMC5662604

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Algorithm for post-clustering curation of DNA amplicon data yields reliable biodiversity estimates By Tobias Guldberg Frøslev, Rasmus Kjøller, Hans Henrik Bruun, Rasmus Ejrnæs, Ane Kirstine Brunbjerg, Carlotta Pietroni, Anders Johannes Hansen Published in: Nature Communications DOI: 10.1038/s41467-017-01312-x ABSTRACT DNA metabarcoding is promising for cost-effective biodiversity monitoring, but reliable diversity estimates are difficult to achieve and validate. Here we present and validate a method, called LULU, for removing erroneous molecular operational taxonomic units (OTUs) from community data derived by high-throughput sequencing of amplified marker genes. LULU identifies errors by combining sequence similarity and co-occurrence patterns. To validate the LULU method, we use a unique data set of high quality survey data of vascular plants paired with plant ITS2 metabarcoding data of DNA extracted from soil from 130 sites in Denmark spanning major environmental gradients. OTU tables are produced with several different OTU definition algorithms and subsequently curated with LULU, and validated against field survey data. LULU curation consistently improves α-diversity estimates and other biodiversity metrics, and does not require a sequence reference database; thus, it represents a promising method for reliable biodiversity estimation. INTRODUCTION Quantifying biodiversity is a key aim of ecological science, but for the majority of organisms, species detection and identification are so demanding and costly that assessment of multi-taxon biodiversity is generally intractable1,2. High-throughput sequencing (HTS) of genetic markers, which have already become standard in microbiology, is a promising tool for rapid, reproducible and thorough censuses of eukaryotic biodiversity in complex ecosystems3–5. However, it is poorly studied whether reliable eukaryotic α-diversity metrics can be achieved from such methods, possibly due to a shortage of comprehensive inventories with paired sets of thorough inventory data and DNA data. Not only are there sampling issues with regard to environmental DNA, but PCR and sequencing processes also generate errors6,7, which, together with intraspecific variation, result in sequence richness far beyond the ‘true’ richness of the sampled biotic community. Given the incomplete reference databases for most organism groups, such errors are not easily separated from true OTU’s. Molecular, ecological and biodiversity studies based on HTS have mainly been developed and applied to microorganisms (bacteria and fungi), for which true diversity generally is poorly known. Such studies have often estimated very high levels of α-diversity4,8,9. These high diversity estimates from HTS data have by some been taken as a first glance into a hitherto un-sampled ‘rare biosphere’9, by others as an argument for deleting the rarest OTUs at some arbitrary level. While many rare OTUs are beyond doubt real biological entities10, an appreciable fraction of rare OTUs are likely errors from PCR and sequencing11. The proportion of erroneous singletons has been estimated to 38% on average12, and studies of bacterial mock communities have revealed that standard bioinformatic approaches result in many spurious OTUs13. Bioinformatic error-reduction has mainly been focussed on selective removal of low-quality reads7,11 and so-called chimeric sequences14,15, recently implemented in model-based pipelines13. Other approaches have focused on laboratory measures to reduce the number of PCR and sequencing errors and the potential impact of these16–18. Although these advances have greatly improved data based on amplicon sequencing, erroneous OTUs remain a critical issue. Most data processing algorithms cluster sequences into OTUs based on similarity (often set to around 97% depending on organism group and marker region) approximately corresponding to recognized average genetic species boundaries. To minimize the number of errors, the tail of the OTU rank-abundance distribution is then often discarded at some arbitrary level (usually only discarding singletons, but often more), assuming that a large proportion of infrequent OTUs represent errors4,11,12,19. However, in real ecosystems, genuinely infrequent species should be expected to make up the lion’s share of total richness. Thus, current approaches are likely to retain only dominant species, which may be adequate for assessing composition and turnover, but much less so for α-diversity. Aiming for improved diversity estimates and a taxonomic composition better aligned with ‘reality’, we developed a co-occurrence based post-clustering curation method, LULU. The LULU algorithm excludes artefactual OTUs without discarding rare, but real OTUs. The core mechanism is the identification and merging of ‘daughter’ OTUs with consistently co-occurring, sequence similar, but more abundant ‘parent’ OTUs across a multi-sample data set, under the assumption that the ‘daughter’ OTUs are artefacts. The algorithm is independent of a reference database, and can thus be applied to any OTU table produced for any set of samples produced by any initial OTU definition algorithm. A related approach—distribution based clustering—was developed to cluster 16S bacterial sequence data into ecologically significant OTUs20, and recently implemented in the dbotu3 tool21. However, whereas LULU is a post-clustering curation method aiming at removing erroneous OTUs to achieve meaningful diversity metrics, dbotu3 is a clustering method aiming at identifying ecologically significant haplotypes of bacterial strains. Despite different objectives and parameters, the overall processing strategy of the dbotu3 algorithm is similar to LULU. To validate the LULU algorithm, we used a plant data set for ITS2 (nuclear ribosomal internal transcribed spacer region 2) obtained from an extensive soil sampling across 130 field sites in Denmark, for which thorough reference data (presence/absence) of vascular plants were obtained concurrently. A total of 564 plant species (approximately one third of the naturally occurring flora) were recorded in the study, with field site species richness per site ranging from 6 to 93. For the main part of the validation, we used the botanical survey data as ground truth data for species-level α-diversity and composition. For additional parts of the validation procedure, we assigned OTUs to species in public reference databases (GenBank), which contained ITS2 data for 88% of species recorded in the field survey, allowing satisfactory taxonomic resolution for most OTUs. OTU tables were produced from the sequence data using a set of representative clustering methods: VSEARCH22, SWARM23, CROP24 and DADA213. All tables were subsequently curated with the LULU algorithm, and the curation effect was evaluated against ground truth in the form of plant survey data. We also tested the dbotu3 algorithm, both for ‘one-step’ clustering as intended, and as an alternative to LULU for post-clustering curation. Furthermore, we evaluated the effect of singleton removal in comparison to curation. Finally, we visualized and evaluated the exact curation effect on selected genera of plants. We show that co-occurrence based post-clustering curation greatly improve diversity measures for all tested OTU tables for a large set of metrics. We conclude that LULU is a tool with far-reaching potential for practical application where realistic biodiversity metrics are needed. RESULTS Curation improves correlation with plant richness OTU richness data for the un-curated and curated tables across the 130 sampling sites were regressed against the corresponding observed plant richness, for each clustering approach and similarity level separately (Fig. 1a (97% clustering level), Table 1, Supplementary Figs 1, 2 (all clustering levels)). Three measures for correspondence with ‘real richness’ were examined: (i) The coefficient of determination (R 2) was used as a measure of goodness of prediction, (ii) an intercept close to zero was expected in combination with (iii) a regression slope close to—but less than—unity to be indicative of a realistic prediction, reflecting systematic soil sampling to be less effective than a thorough botanical survey, although DNA from soil for several reasons may in fact contain DNA from more species

Read the complete text