BiLinT

BiLinT: Bayesian inference of lineage tree by joint analysis of multi-modal lineage tracing data

Software dependencies

BiLinT is written with R and C++. Before implementing our software, please install the following packages in R:

tidyr, dplyr, coda, Rcpp, RcppArmadillo, ggplot2, PCMBase, ape, ggtree, castor, phangorn, TreeSearch, TreePar, TreeSim, msm, Rfast, stringr, treeio, TreePar, ggstance, TreeTools, phytools, PCMBaseCpp, rdist, foreach, doParallel, Rcpp2doParallel

Usage

To use BiLinT, please set R working directory to BiLinT-master after downloading this repository. Please make sure you have installed all the dependencies correctly, and then open source code BiLinT_app.R to execute the commands line by line as following.

• In Model Input section, first set random seed myseed, and then specify the folder foldername to save output files (a new folder will be created if it does not exist). For example:

  #################################################################
  ###################### MODEL INPUT ##############################
  #################################################################
  
  myseed <-  1               # set random seed
  foldername <-  "temp_out"          # set output foldername
  dataname <- 'toy_model'
  dir.create(foldername)  # folder where outputs are saved
  

  Then input the barcode matrix, seq_data_mt, the gene expression matrix obs_traits_original and the scaring window for barocde data, including t1, t2 and t_tol. For the given example data:

  muti_data <- readRDS('toy_model_data.rds')
  seq_data_mt <- muti_data$seq_data_mt # barcode data, where row represents cells, columns represent targets
  M <- dim(muti_data$seq_data_mt)[1]
  N <- dim(muti_data$seq_data_mt)[2]
  
  obs_traits_original <- muti_data$obs_traits # normalized gene expression data, where row represents cells, columns represent genes
  obs_traits <- prcomp(obs_traits_original, rank. = 5)$x
  K <- dim(obs_traits)[2]
  state <- sort(setdiff(unique(as.vector(seq_data_mt)), c(0, -1)))
  
  t1 <- 0 # scaring start time
  t2 <- muti_data$seq_para_data$t2 # scaring end time
  t_tol <- muti_data$seq_para_data$t_tol # total experiment time
  

• Next, assign Bayesian sampling parameters in specify_pars.R. For most of the parameters, BiLinT works just fine with default values. Some of the parameters you can change are:

  MCMC_par$burnin <- 500  # burnin sample size
  MCMC_par$Nsamp <- 500   # number of samples for inference
  

• After that, execute sampler.R to perform MCMC sampling, then the samples used for inference are stored in the .Rdata files;

• Finally, get the tree estimation by

  tree_est <- tree_estimate(Trace)
  

• The tree can be displayed by

  ggtree(tree_est, size = 1,  layout = "circular",branch.length = "none") +
  geom_tiplab(size=5) +
  geom_point2(aes(subset=(node %in% c(65:127))),size=2)+
  theme_tree()+ theme(legend.position = "none")