occumb {occumb}

Model-fitting function.

Description

occumb() fits the multispecies site-occupancy model for eDNA metabarcoding (Fukaya et al. 2022) and returns a model-fit object containing posterior samples.

Usage

occumb(
  formula_phi = ~1,
  formula_theta = ~1,
  formula_psi = ~1,
  formula_phi_shared = ~1,
  formula_theta_shared = ~1,
  formula_psi_shared = ~1,
  prior_prec = 1e-04,
  prior_ulim = 10000,
  data,
  n.chains = 4,
  n.adapt = NULL,
  n.burnin = 10000,
  n.thin = 10,
  n.iter = 20000,
  parallel = FALSE,
  ...
)

Arguments

formula_phi

A right-hand side formula describing species-specific effects of sequence relative dominance (\phi).

formula_theta

A right-hand side formula describing species-specific effects of sequence capture probability (\theta).

formula_psi

A right-hand side formula describing species-specific effects of occupancy probability (\psi).

formula_phi_shared

A right-hand side formula describing effects of sequence relative dominance (\phi) that are common across species. The intercept term is ignored (see Details).

formula_theta_shared

A right-hand side formula describing effects of sequence capture probability (\theta) that are common across species. The intercept term is ignored (see Details).

formula_psi_shared

A right-hand side formula describing effects of occupancy probability (\psi) that are common across species. The intercept term is ignored (see Details).

prior_prec

Precision of normal prior distribution for the community-level average of species-specific parameters and effects common across species.

prior_ulim

Upper limit of uniform prior distribution for the standard deviation of species-specific parameters.

data

A dataset supplied as an occumbData class object.

n.chains

Number of Markov chains to run.

n.adapt

Number of iterations to run in the JAGS adaptive phase.

n.burnin

Number of iterations at the beginning of the chain to discard.

n.thin

Thinning rate. Must be a positive integer.

n.iter

Total number of iterations per chain (including burn-in).

parallel

If TRUE, run MCMC chains in parallel on multiple CPU cores.

...

Additional arguments passed to jags() function.

Details

occumb() allows the fitting of a range of multispecies site occupancy models, including covariates at different levels of the data generation process. The most general form of the model can be written as follows (the notation follows that of the original article; see References).

Sequence read counts:

(y_{1jk}, ..., y_{Ijk}) \sim \textrm{Multinomial}((\pi_{1jk}, ..., \pi_{Ijk}), N_{jk}),

\pi_{ijk} = \frac{u_{ijk}r_{ijk}}{\sum_m u_{mjk}r_{mjk}},

Relative frequency of species sequences:

r_{ijk} \sim \textrm{Gamma}(\phi_{ijk}, 1),

Capture of species sequences:

u_{ijk} \sim \textrm{Bernoulli}(z_{ij}\theta_{ijk}),

Site occupancy of species:

z_{ij} \sim \textrm{Bernoulli}(\psi_{ij}),

where the variations of \phi, \theta, and \psi are modeled by specifying model formulas in formula_phi, formula_theta, formula_psi, formula_phi_shared, formula_theta_shared, and formula_psi_shared.

Each parameter may have species-specific effects and effects that are common across species, where the former is specified by formula_phi, formula_theta, and formula_psi, whereas formula_phi_shared, formula_theta_shared, and formula_psi_shared specify the latter. As species-specific intercepts are specified by default, the intercept terms in formula_phi_shared, formula_theta_shared, and formula_psi_shared are always ignored. Covariate terms must be found in the names of the list elements stored in the spec_cov, site_cov, or repl_cov slots in the dataset object provided with the data argument. Covariates are modeled using the log link function for \phi and logit link function for \theta and \psi.

The two arguments, prior_prec and prior_ulim, control the prior distribution of parameters. For the community-level average of species-specific effects and effects common across species, a normal prior distribution with a mean of 0 and precision (i.e., the inverse of the variance) prior_prec is specified. For the standard deviation of species-specific effects, a uniform prior distribution with a lower limit of zero and an upper limit of prior_ulim is specified. For the correlation coefficient of species-specific effects, a uniform prior distribution in the range of -1 to 1 is specified by default.

See the package vignette for details on the model specifications in occumb().

The data argument requires a dataset object to be generated using ocumbData(); see the document of occumbData().

The model is fit using the jags() function of the jagsUI package, where Markov chain Monte Carlo methods are used to obtain posterior samples of the parameters and latent variables. Arguments n.chains, n.adapt, n.burnin, n.thin, n.iter, and parallel are passed on to arguments of the same name in the jags() function. See the document of jagsUI's jags() function for details.

Value

An S4 object of the occumbFit class containing the results of the model fitting and the supplied dataset.

References

K. Fukaya, N. I. Kondo, S. S. Matsuzaki and T. Kadoya (2022) Multispecies site occupancy modelling and study design for spatially replicated environmental DNA metabarcoding. Methods in Ecology and Evolution 13:183–193. doi:10.1111/2041-210X.13732

Examples

# Generate the smallest random dataset (2 species * 2 sites * 2 reps)
I <- 2 # Number of species
J <- 2 # Number of sites
K <- 2 # Number of replicates
data <- occumbData(
    y = array(sample.int(I * J * K), dim = c(I, J, K)),
    spec_cov = list(cov1 = rnorm(I)),
    site_cov = list(cov2 = rnorm(J),
                    cov3 = factor(1:J)),
    repl_cov = list(cov4 = matrix(rnorm(J * K), J, K)))


# Fitting a null model (includes only species-specific intercepts)
res0 <- occumb(data = data)

# Add species-specific effects of site covariates in occupancy probabilities
res1 <- occumb(formula_psi = ~ cov2, data = data)        # Continuous covariate
res2 <- occumb(formula_psi = ~ cov3, data = data)        # Categorical covariate
res3 <- occumb(formula_psi = ~ cov2 * cov3, data = data) # Interaction

# Add species covariate in the three parameters
# Note that species covariates are modeled as common effects
res4 <- occumb(formula_phi_shared = ~ cov1, data = data)   # phi
res5 <- occumb(formula_theta_shared = ~ cov1, data = data) # theta
res6 <- occumb(formula_psi_shared = ~ cov1, data = data)   # psi

# Add replicate covariates
# Note that replicate covariates can only be specified for theta and phi
res7 <- occumb(formula_phi = ~ cov4, data = data)   # phi
res8 <- occumb(formula_theta = ~ cov4, data = data) # theta

# Specify the prior distribution and MCMC settings explicitly
res9 <- occumb(data = data, prior_prec = 1E-2, prior_ulim = 1E2,
               n.chains = 1, n.burnin = 1000, n.thin = 1, n.iter = 2000)
res10 <- occumb(data = data, parallel = TRUE, n.cores = 2) # Run MCMC in parallel