An Rshiny app for modelling environmental DNA data: accounting for false positive and false negative observation error

Abstract

Environmental DNA (eDNA) surveys have become a popular tool for assessing the distribution of species. However, it is known that false positive and false negative observation error can occur at both stages of eDNA surveys, namely the field sampling stage and laboratory analysis stage.We present an RShiny app that implements the Griffin et al. (2019) statistical method, which accounts for false positive and false negative errors in both stages of eDNA surveys. Following Griffin et al. (2019), we employ a Bayesian approach and perform efficient Bayesian variable selection to identify important predictors for the probability of species presence as well as the probabilities of observation error at either stage.We demonstrate the RShiny app using a data set on great crested newts collected by Natural England in 2018 and we identify water quality, pond area, fish presence, macrophyte cover, frequency of drying as important predictors for species presence at a site.The state-of-the-art statistical method that we have implemented is the only one that has specifically been developed for the purposes of modelling false negatives and false positives in eDNA data. Our RShiny app is user-friendly, requires no prior knowledge of R and fits the models very efficiently. Therefore, it should be part of the tool-kit of any researcher or practitioner who is collecting or analysing eDNA data.

Integrating eDNA metabarcoding and simultaneous underwater visual surveys to describe complex fish communities in a marine biodiversity hotspot

“…We used the single-species, single-season occupancy model proposed by Griffin et al, (2020) using the freely available eDNA Shiny app (https://seak.shiny apps.io/eDNA/) (Diana et al, 2020). This model allows estimating multiple parameters, including the probability that a site is occupied by a species (occupancy, ѱ), the probability of eDNA capture in the environmental samples (capture, θ), and the probability of detection of a species in PCR assays (detectability, p) (Ficetola et al, 2014;Griffin et al, 2020;Guillera-Arroita et al, 2017;Lahoz-Monfort et al, 2016;Schmidt et al, 2013).…”

Reliability of environmental DNA surveys to detect pond occupancy by newts at a national scale

“…The posterior mean estimates of 0.198 for occupancy are comparable to most other estimates for great crested newt pond occupancy in the UK, but lower than the naïve estimate. The lower modelled estimates of occupancy than the naïve estimate suggest that false positives should not be ignored and need to be accounted for statistically using methodologies such as the eDNAShinyApp package used here 23 , 24 , 27 .…”

“…We would recommend that, where possible, results from individual sites are interpreted as a probability of site occupancy, based on modelled outputs such as those produced by the eDNAShinyApp R package 23 , 27 . The precision of these models is dependent on sample size.…”

“…Secondly, it provides opportunities to explore how occupancy and detection rates are influenced by habitat covariates. Although occupancy modelling has been previously applied to great crested newt eDNA data, the examples have used either localised data 22 , analysis using standard occupancy models 22 or have been presented in the literature in a methodological or model development context 21 , 23 , 24 . As a result, outputs from national data sets have yet to be examined in an ecological context that relate to great crested newt eDNA surveys.…”

“…Using an eDNA data set collected from across England, we use the Griffin et al (2020) model 21 via the eDNAShinyApp package 23 , within statistical software R 27 , to examine pond occupancy and false positive and false negative error rates of surveys targeting great crested newt eDNA. This allows us to estimate national pond great crested newt occupancy rates and to consider habitat covariates collected alongside the eDNA samples, or gleaned from location data, to relate pond occupancy to site characteristics.…”