Navigating the seven challenges of taxonomic reference databases in metabarcoding analyses

Abstract

Assessment of biodiversity using metabarcoding data, such as from bulk or environmental DNA sampling, is becoming increasingly relevant in ecology, biodiversity sciences and monitoring. Thereby, the taxonomic identification of species from their DNA sequences relies strongly on reference databases that link genetic sequences to taxonomic names. These databases vary in completeness and availability, depending on the taxonomic group studied and the genetic region targeted. The incompleteness of reference databases is an important argument to explain the nondetection by metabarcoding of species supposedly present. However, there exist further and generally overlooked problems with reference databases that can lead to false or inaccurate inferences of taxonomic assignment. Here, we synthesize all possible problems inherent to reference databases. In particular, we identify a complete, mutually nonexclusive list of seven classes of challenges when it comes to selecting, developing and using a reference database for taxonomic assignment. These are: (i) mislabelling, (ii) sequencing errors, (iii) sequence conflict, (iv) taxonomic conflict, (v) low taxonomic resolution, (vi) missing taxa and (vii) missing intraspecific variants. For each problem identified, we provide a description of possible consequences on the taxonomic assignment process. We illustrate the respective problem with examples taken from the literature or obtained by quantitative analyses of public databases, such as GenBank or BOLD. Finally, we discuss possible solutions to the identified problems and how to navigate them. Only by raising users' awareness of the limitations of metabarcoding data and DNA reference databases will adequate interpretations of these data be achieved.

A combination of machine‐learning and eDNA reveals the genetic signature of environmental change at the landscape levels

“…For example, OTU 142_141 is unclassified but plays an important role for the DNA model performance (Figure 4). Several reasons can explain the failure of the classification of these sequences, such as incompleteness or conflicts in the reference database (Keck et al, 2023) but these data remain nonetheless valuable and machine learning algorithms can be used to incorporate them. We here highlight the value of using complete metabarcoding data for the classification and assessment of ecological state, and advocate approaches that are not perpetuating previous limitations inherent to a past method yet obsolete to novel approaches.…”

“…Several reasons can explain the failure of the classification of these sequences, such as incompleteness or conflicts in the reference database (Keck et al, 2023) but these data remain nonetheless valuable and machine learning algorithms can be used to incorporate them.…”

“…However, some inherent limits remain, because metabarcoding data are not fully compatible with the taxonomic classifications in use, and the attempt to reproduce existing indices by novel means retained constraints of the old methods. As a major limitation and side‐effect of this “reproduce” yet not “extend” strategy, a large part of the genetic data produced remains unassignable, due to the use of biased and non‐specific primers (Casey et al, 2021; Clarke et al, 2017; Ficetola et al, 2021) and limited reference databases (Keck et al, 2023; Li et al, 2022; Weigand et al, 2019). As a result, estimates of community composition obtained by traditional methods and DNA metabarcoding are not always congruent (Keck et al, 2022).…”

Opportunities and inherent limits of using environmental DNA for population genetics

“…Linking data to a particular taxon is thus highly dependent on the availability of reliable reference sequences, encompassing the whole range of genetic variability. The completeness and accuracy of reference databases is a well‐known challenge for barcoding and metabarcoding studies, as described in Keck et al (2023). Although some methods exist to curate reference databases and remove erroneous sequences, the extant of problematic or missing references is usually difficult to evaluate.…”

“…Linking data to a particular taxon is thus highly dependent on the availability of reliable reference sequences, encompassing the whole range of genetic variability. The completeness and accuracy of reference databases is a well-known challenge for barcoding and metabarcoding studies, as described inKeck et al (2023). Although some methods exist to curate reference databases and remove erroneous sequences, the extant of problematic or missing references is usually difficult to evaluate.Unlike the absence of references for a particular species, the absence of particularly divergent genetic variants of a target species F I G U R E 3 Representation of the differences between the genetic information retrieved with either traditional individual-based or eDNA sampling in two fish populations of the same river.…”