Usage
bevimed(y, G, min_ac = 1L, tau_shape = c(1, 1), pi_shape = c(6, 1),
omega_shape = if (max(min_ac) == 1L) c(2, 8) else c(2, 2),
samples_per_chain = 1000, stop_early = FALSE, blocks = 5,
burn = as.integer(samples_per_chain/10), temperatures = (0:6/6)^2,
tune_temps = 0, return_z_trace = TRUE, return_x_trace = FALSE,
swaps = as.integer(length(temperatures)/2), optimise_z0 = FALSE,
tune_omega_and_phi_proposal_sd = FALSE, tune_block_size = 100,
variant_weights = NULL, standardise_weights = TRUE,
log_phi_mean = -0.15, log_phi_sd = sqrt(0.3),
tandem_variant_updates = if (max(min_ac) == 1) 0 else min(sum(y), nrow(G)),
...)
Arguments
y
Logical vector of case (TRUE) control (FALSE) status.
G
Integer matrix of variant counts per individual, one row per individual and one column per variant.
min_ac
Minimum allele count per individual to be considered to have a pathogenic combination of variants. '1' could correspond to a dominant inheritance hypothesis whereas '2' could correspond to a recessive inheritance hypothesis. Can either be an integer value (then used for all samples), or integer vector of length(y), useful for example when different samples have different ploidy.
tau_shape
Beta shape hyper-priors for prior on rate of affection (i.e. being a case) amongst individuals with non-pathogenic variant combinations (i.e. they have less than min_ac variants.
pi_shape
Beta shape hyper-priors for prior on rate of affection (i.e. being a case) amongst individuals with pathogenic variant combinations (i.e. they have at least min_ac variants.
omega_shape
Beta shape hyper-priors for prior on rate of pathogenicity amongst variants.
samples_per_chain
Number of samples to draw from each chain.
stop_early
Logical value determining whether to attempt to stop the sampling as soon as certain conditions are met (i.e. either the estimated marginal log likelihood lies within a certain confidence interval, or we are sufficiently confidence that the log Bayes factor against of model gamma = 1 over model gamma = 0 is sufficiently low).
blocks
Maximum number of blocks of length samples_per_chain samples to draw before either the confidence interval for the marginal likelihood under the model gamma = 1 is sufficiently small or terminating the sampling.
burn
Number of samples to drop from the start of the chain.
temperatures
Numeric vector of temperatures of power posteriors. One chain will be created for each element of the vector at the corresponding temperature.
tune_temps
Integer value - if greater than 0, the temperatures argument is ignored, and instead tune_temps tuned temperatures are used instead.
return_z_trace
Logical value determining whether to store the z-vectors for each chain, which uses alot of memory, particularly if samples_per_chain, k and length(temperatures) are large.
return_x_trace
Logical value determining whether to store the x variable determined by success samples of z. Potentially uses alot of memory, particularly if samples_per_chain, k and length(temperatures) are large.
swaps
Number of swaps between adjacent tempered chains to perform per update cycle.
optimise_z0
Logical value determining whether to use a simulated annealing optimisation run to tune the initial values of z.
tune_omega_and_phi_proposal_sd
Logical value determining whether the proposal SDs of the Metropolis-Hastings estimated parameters should be tuned for a target acceptance range.
tune_block_size
Integer value giving number of samples to draw when estimatating the acceptance rate of the omega/phi proposals.
variant_weights
Vector of log-odds off-sets for rates of pathogenicity of individual variants relative to the global rate, omega.
standardise_weights
Boolean value determining whether weights should be standardised by subtracting their mean and dividing by their sample standard deviation. If FALSE, weights are untransformed.
log_phi_mean
Mean for normal prior on scaling factor phi.
log_phi_sd
SD for normal prior on scaling factor phi. Setting to 0 causes the weights to be fixed and not estimated.
tandem_variant_updates
Number of tandem variant updates to make per update cycle.