module Model where import Probability import Probability.Logger import Bio.Alphabet import Bio.Alignment import Bio.Sequence import MCMC (scaleGroupSlice) import Tree import Tree.Newick import SModel import System.Environment (getArgs) import System.FilePath {- Problems: 1. We can explain the data ignoring the sample ages if we have high low mutation rates and deep coalescence times. - We want the prior to favor high mutation rates. - We want the prior to favor a recent root age => high coalescent rates => small population sizes. 3. N/tau prints as tau -} smodel_prior nucleotides = do freqs <- sample $ symmetricDirichletOn (getLetters nucleotides) 1 kappa1 <- sample $ logNormal (log 2) (1/4) kappa2 <- sample $ logNormal (log 2) (1/4) let tn93_model = tn93' nucleotides kappa1 kappa2 freqs let loggers = ["kappa1" %=% kappa1, "kappa2" %=% kappa2, "frequencies" %=% freqs] return (tn93_model, loggers) tree_prior taxa = do popSize <- sample $ logNormal 3 2 -- Favor small population sizes -> high coalescent rates -> more recent root age. let taxonAges = getTaxonAges taxa "s(\\d+)$" Forward popSizes = [(0, popSize)] tree <- sample $ coalescentTree taxonAges popSizes let loggers = ["N_over_tau" %=% popSize] return (tree, loggers) model seqData@[seqData1,seqData2,seqData3] logTree = do let taxa = getTaxa seqData1 (tree , tree_loggers) <- tree_prior taxa (smodel, sloggers ) <- smodel_prior dna mu <- sample $ logNormal (-5) 1.25 -- Favor high mutation rates -> more recent root age. -- We can't inverse-scale mu because it is exp(modifiable), not directly modifiable. addMove 1 $ scaleGroupSlice [ nodeTime tree node | node <- internalNodes tree ] weights <- map (3*) <$> (sample $ symmetricDirichlet 3 3) let [mu1,mu2,mu3] = map (mu*) weights observe seqData1 $ phyloCTMC tree (alignmentLength seqData1) smodel mu1 observe seqData2 $ phyloCTMC tree (alignmentLength seqData2) smodel mu2 observe seqData3 $ phyloCTMC tree (alignmentLength seqData3) smodel mu3 addLogger $ logTree (addInternalLabels tree) let tlength = treeLength tree substs = map (parsimony tree (unitCostMatrix dna)) seqData rootAge = nodeTime tree (root tree) loggers = ["tree" %>% tree_loggers, "tn93" %>% sloggers, "|T|" %=% tlength, "mu*|T|" %=% tlength * mu, "rootAge" %=% rootAge, "weights" %=% weights, "mu" %=% mu, "mu1" %=% mu1, "mu2" %=% mu2, "mu3" %=% mu3, "#substs" %=% substs, "#total_substs" %=% sum substs] return loggers main logDir = do [filename] <- getArgs sequences <- loadSequences filename let seqData = [ mkAlignedCharacterData dna $ selectRange range sequences | range <- ["3-.\\3", "1-.\\3", "2-.\\3"]] logTree <- treeLogger $ logDir "C1.trees" return $ model seqData logTree