Rev Language Reference

dnNucleotideRateMatrixMixture(RealPos[] alpha, String fixedModel, Real lnPriorOdds)

alpha :	RealPos[] (pass by const reference)
	The alpha shape parameter.
fixedModel :	String (pass by value)
	Confine the chain to one model: "free" (the default) lets the reversible jump operate, while "reversible" or "nonreversible" starts the chain in that model and suppresses the jump moves entirely. Use this to measure the two models separately.
	Default : free
lnPriorOdds :	Real (pass by value)
	The prior odds on the model indicator, as log(rho_R/rho_N), where rho_R and rho_N are the prior probabilities of the time-reversible and non-reversible models. The default of zero gives the two models equal prior probability. A positive value favours the time-reversible model. The odds are held fixed unless the move is given tuneModelPrior=TRUE, in which case they are adapted during burn-in and then frozen.
	Default : 0

RateGenerator

Use this distribution with `mvMPQRateMatrix` to sample the rate matrix and propose changes between the two models. `alpha` is a vector of four positive Dirichlet concentration parameters for the stationary nucleotide frequencies. Setting all four entries to one gives a uniform prior on these frequencies. `fixedModel` defaults to `"free"`, allowing transitions between models. Set it to `"reversible"` or `"nonreversible"` to restrict the analysis to that model; model-switching proposals are then disabled. `lnPriorOdds` specifies log(rho_R/rho_N), where rho_R and rho_N are the prior probabilities of the reversible and non-reversible models. Its default, zero, gives equal prior probabilities. Positive values favor reversibility; negative values favor non-reversibility. For a free-model analysis with fixed prior odds, the Bayes factor favoring the non-reversible model is BF_NR = (p_N / p_R) * exp(lnPriorOdds) where p_N and p_R are the posterior model probabilities estimated from the sampled model indicators. Use production samples after burn-in, with adequate mixing between models. If `mvMPQRateMatrix` adapts the model prior during burn-in, freeze that adaptation before collecting production samples and use the final value reported by `Q.lnPriorOdds()` in the calculation.

# Create a rate matrix with equal prior probabilities for the models.
Q ~ dnNucleotideRateMatrixMixture(alpha=v(1,1,1,1), lnPriorOdds=0.0)

moves = VectorMoves()
moves.append(mvMPQRateMatrix(Q, tune=FALSE, tuneModelPrior=FALSE))

# Define quantities that can be monitored during MCMC.
reversible := ifelse(Q.isReversible(), 1.0, 0.0)
frequencies := Q.getPi()
log_prior_odds := Q.lnPriorOdds()

# Alternatively, restrict a separate matrix to one model.
Q_NR ~ dnNucleotideRateMatrixMixture(alpha=v(1,1,1,1), fixedModel="nonreversible")

• mvMPQRateMatrix
• mvMPQTree
• mvReversibilityAwareBranchLengthScale
• dnPhyloCTMC