6 Example workflows
-Three example workflows will be discussed in this vignette: (1) Inferring model without an initial model, (2) Inferring model with an initial model, (3) Extending model with more genes. The two workflows are largely similar, which only differ in the data preparation step.
-6.1 Inferring model without an initial model
-This workflow is intended for use on inferring a Boolean model without an initial model.
-When no initial model is used, BoolTraineR will reconstruct gene interactions from a list of user-specified genes. If the number of genes in the expression data is low (e.g. in qPCR), it is also possible to use all the genes in the expression data.
-6.1.1 Full workflow
-Full workflow is included here for easy referencing. Each step is discussed in further details below.
-set.seed(0) #use to ensure reproducibility. remove in actual use.
-
-# (1) Setup paths and environment.
-library(BoolTraineR)
-
-# If intending to use parallel processing, uncomment the following lines.
-# library(doParallel) num_core = 4 #specify the number of cores to be used.
-# doParallel::registerDoParallel(cores=num_core)
-
-# (2) Load data.
-data(wilson_raw_data) #load a data frame of expression data.
-tmp_data = initialise_raw_data(wilson_raw_data, max_expr = "low")
-cdata = tmp_data[[1]] #continuous data
-ddata = tmp_data[[2]] #discretised data
-
-# (3) Filter cell types.
-cell_ind = grepl("cmp", rownames(cdata)) | grepl("gmp", rownames(cdata)) | grepl("mep",
- rownames(cdata))
-fcdata = cdata[cell_ind, ] #select only relevant cells.
-fddata = ddata[cell_ind, ]
-
-# (4) Filter genes.
-gene_ind = c("fli1", "gata1", "gata2", "gfi1", "scl", "sfpi1") #select genes to be included.
-fcdata = fcdata[, gene_ind]
-fddata = fddata[, gene_ind]
-
-# (5) Inferring Boolean model.
-final_model = model_train(cdata = fcdata, ddata = fddata, max_varperrule = 4,
- verbose = T)
-
-# (6) Visualise the Boolean model generated.
-plotBM(final_model)6.1.2 Initial setup
-The first step is to load the BoolTraineR package. If you are intending to use parallel processing, you will also need to load the doParallel package. Then specify how many cores you intend to use using registerDoParallel from the doParallel package.
set.seed(0) #use to ensure reproducibility. remove in actual use.
-
-# (1) Setup paths and environment.
-library(BoolTraineR)
-
-# If intending to use parallel processing, uncomment the following lines.
-# library(doParallel) num_core = 4 #specify the number of cores to be used.
-# doParallel::registerDoParallel(cores=num_core)6.1.3 Data preparation
-Only the expression data is needed for inferring a Boolean model without an initial model.
-To load the data into R, use read.table or read.csv. In this example, we are using the example data included with the package, so we are accessing it by using data.
initialise_raw_data is used to preprocess the data.
# (2) Load data.
-data(wilson_raw_data) #load a data frame of expression data.
-tmp_data = initialise_raw_data(wilson_raw_data, max_expr = "low")
-cdata = tmp_data[[1]] #continuous data
-ddata = tmp_data[[2]] #discretised dataOnce data is loaded and preprocessed, filter the cell types or genes to be included in the analysis if needed. It is advisable to reduce the number of genes to be included if the computation takes too long to complete.
-# (3) Filter cell types.
-cell_ind = grepl("cmp", rownames(cdata)) | grepl("gmp", rownames(cdata)) | grepl("mep",
- rownames(cdata))
-fcdata = cdata[cell_ind, ] #select only relevant cells.
-fddata = ddata[cell_ind, ]
-
-# (4) Filter genes.
-gene_ind = c("fli1", "gata1", "gata2", "gfi1", "scl", "sfpi1") #select genes to be included.
-fcdata = fcdata[, gene_ind]
-fddata = fddata[, gene_ind]6.1.4 Run model training
-To reconstruct a Boolean model from an expression data, run model_train.
In this example, model_train takes a few seconds to be completed on a single core. If this steps take a very long time to complete, do consider using the parallel processing option as described above.
You will receive a BoolModel object at the end of the model training process. The BoolModel object can be visualise quickly using plotBM, which is based on igraph package. For easier manipulation, output the Boolean model using outgraph_model and display it with Cytoscape or Gephi.
# (5) Inferring Boolean model.
-final_model = model_train(cdata = fcdata, ddata = fddata, max_varperrule = 4,
- verbose = T)
-
-# (6) Visualise the Boolean model generated.
-plotBM(final_model)
6.2 Inferring model with an initial model
-This workflow is intended for use on inferring a Boolean model with an initial model.
-When an initial model is used, note that only genes that are both present in the initial model and expression data will be used for reconstructing gene interactions. Any genes in the initial model that do not have corresponding expression values in the data will keep their original gene interactions as specified in the initial model without any modifications.
-6.2.1 Full workflow
-Full workflow is included here for easy referencing. Each step is discussed in further details below.
-set.seed(0) #use to ensure reproducibility. remove in actual use.
-
-# (1) Setup paths and environment.
-library(BoolTraineR)
-
-# If intending to use parallel processing, uncomment the following lines.
-# library(doParallel) num_core = 4 #specify the number of cores to be used.
-# doParallel::registerDoParallel(cores=num_core)
-
-# (2) Load data.
-data(krum_bmodel) #load a data frame of Boolean model.
-data(krum_istate) #load a data frame of initial state.
-data(wilson_raw_data) #load a data frame of expression data.
-
-bmodel = initialise_model(krum_bmodel)
-istate = krum_istate
-tmp_data = initialise_raw_data(wilson_raw_data, max_expr = "low")
-cdata = tmp_data[[1]] #continuous data
-ddata = tmp_data[[2]] #discretised data
-
-# (3) Filter cell types.
-cell_ind = grepl("cmp", rownames(cdata)) | grepl("gmp", rownames(cdata)) | grepl("mep",
- rownames(cdata))
-fcdata = cdata[cell_ind, ] #select only relevant cells.
-fddata = ddata[cell_ind, ]
-
-# (4) Inferring Boolean model.
-final_model = model_train(cdata = fcdata, ddata = fddata, bmodel = bmodel, istate = istate,
- max_varperrule = 4, verbose = T)
-
-# (5) Visualise the Boolean model generated.
-plotBM(final_model)6.2.2 Initial setup
-The first step is to load the BoolTraineR package. If you are intending to use parallel processing, you will also need to load the doParallel package. Then specify how many cores you intend to use using registerDoParallel from the doParallel package.
set.seed(0) #use to ensure reproducibility. remove in actual use.
-
-# (1) Setup paths and environment.
-library(BoolTraineR)
-
-# If intending to use parallel processing, uncomment the following lines.
-# library(doParallel) num_core = 4 #specify the number of cores to be used.
-# doParallel::registerDoParallel(cores=num_core)6.2.3 Data preparation
-3 pieces of data are needed to infer a Boolean model with an initial model: an expression data, an initial Boolean model and an initial state.
-To load the data into R, use read.table or read.csv. In this example, we are using the example data included with the package, so we are accessing it by using data.
initialise_model converts the data frame containing the Boolean model into a BoolModel object. initialise_raw_data is used to preprocess the data.
# (2) Load data. (2) Load data.
-data(krum_bmodel) #load a data frame of Boolean model.
-data(krum_istate) #load a data frame of initial state.
-data(wilson_raw_data) #load a data frame of expression data.
-
-bmodel = initialise_model(krum_bmodel)
-istate = krum_istate
-tmp_data = initialise_raw_data(wilson_raw_data, max_expr = "low")
-cdata = tmp_data[[1]] #continuous data
-ddata = tmp_data[[2]] #discretised dataOnce data are loaded and preprocessed, filter the cell types or genes to be included in the analysis if needed. It is advisable to reduce the number of genes to be included if the computation takes too long to complete. In this example, genes are not filtered as all genes that are present in both expression data and Boolean model are used automatically.
-# (3) Filter cell types.
-cell_ind = grepl("cmp", rownames(cdata)) | grepl("gmp", rownames(cdata)) | grepl("mep",
- rownames(cdata))
-fcdata = cdata[cell_ind, ] #select only relevant cells.
-fddata = ddata[cell_ind, ]6.2.4 Run model training
-To reconstruct a Boolean model from an expression data, run model_train.
In this example, model_train takes one or two minutes to be completed on a single core. If this steps take a very long time to complete, do consider using the parallel processing option as described above.
You will receive a BoolModel object at the end of the model training process. The BoolModel object can be visualise using plotBM, which is based on igraph package. For easier manipulation, output the Boolean model using outgraph_model and display it with Cytoscape or Gephi.
# (4) Inferring Boolean model.
-final_model = model_train(cdata = fcdata, ddata = fddata, bmodel = bmodel, istate = istate,
- max_varperrule = 4, verbose = T)
-
-# (5) Visualise the Boolean model generated.
-plotBM(final_model)
6.3 Extending model with more genes
-This workflow is intended for use on extending an initial Boolean model with additional genes.
-When an initial model is used, note that only genes that are both present in the initial model and expression data will be used for reconstructing gene interactions. Any genes in the initial model that do not have corresponding expression values in the data will keep their original gene interactions as specified in the initial model without any modifications.
-6.3.1 Full workflow
-Full workflow is included here for easy referencing. Each step is discussed in further details below.
-Note that this example takes a few minutes to run on a single core. The use of parallel processing is recommended.
-set.seed(0) #use to ensure reproducibility. remove in actual use.
-
-# (1) Setup paths and environment.
-library(BoolTraineR)
-
-# If intending to use parallel processing, uncomment the following lines.
-# library(doParallel) num_core = 4 #specify the number of cores to be used.
-# doParallel::registerDoParallel(cores=num_core)
-
-# (2) Load data.
-data(krum_bmodel) #load a data frame of Boolean model.
-data(krum_istate) #load a data frame of initial state.
-data(wilson_raw_data) #load a data frame of expression data.
-
-bmodel = initialise_model(krum_bmodel)
-istate = krum_istate
-tmp_data = initialise_raw_data(wilson_raw_data, max_expr = "low")
-cdata = tmp_data[[1]] #continuous data
-ddata = tmp_data[[2]] #discretised data
-
-# (3) Filter cell types.
-cell_ind = grepl("cmp", rownames(cdata)) | grepl("gmp", rownames(cdata)) | grepl("mep",
- rownames(cdata))
-fcdata = cdata[cell_ind, ] #select only relevant cells.
-fddata = ddata[cell_ind, ]
-
-# (4) Adding extra genes to the initial Boolean model. extra_genes =
-# setdiff(colnames(wilson_raw_data), bmodel@target) #to view available genes
-# to be added. print(extra_genes) #to view available genes to be added.
-add_gene = "ldb1" #genes to be added: ldb1
-grown_bmodel = grow_bmodel(add_gene, bmodel)
-
-# (5) Estimating initial state for the extra genes. (estimating from CMPs)
-tmp_istate = mean(cdata[grepl("cmp", rownames(cdata)), add_gene])
-tmp_istate = matrix(round(tmp_istate), nrow = 1)
-colnames(tmp_istate) = add_gene
-grown_istate = cbind(istate, tmp_istate)
-grown_istate = initialise_data(grown_istate)
-
-# (6) Inferring Boolean model.
-final_model = model_train(cdata = fcdata, ddata = fddata, bmodel = grown_bmodel,
- istate = grown_istate, verbose = T)
-
-# (7) Visualise the Boolean model generated.
-plotBM(final_model)6.3.2 Initial setup
-The first step is to load the BoolTraineR package. If you are intending to use parallel processing, you will also need to load the doParallel package. Then specify how many cores you intend to use using registerDoParallel from the doParallel package.
set.seed(0) #use to ensure reproducibility. remove in actual use.
-
-# (1) Setup paths and environment.
-library(BoolTraineR)
-
-# If intending to use parallel processing, uncomment the following lines.
-# library(doParallel) num_core = 4 #specify the number of cores to be used.
-# doParallel::registerDoParallel(cores=num_core)6.3.3 Data preparation
-3 pieces of data are needed to infer a Boolean model with an initial model: an expression data, an initial Boolean model and an initial state.
-To load the data into R, use read.table or read.csv. In this example, we are using the example data included with the package, so we are accessing it by using data.
initialise_model converts the data frame containing the Boolean model into a BoolModel object. initialise_raw_data is used to preprocess the data.
# (2) Load data.
-data(krum_bmodel) #load a data frame of Boolean model.
-data(krum_istate) #load a data frame of initial state.
-data(wilson_raw_data) #load a data frame of expression data.
-
-bmodel = initialise_model(krum_bmodel)
-istate = krum_istate
-tmp_data = initialise_raw_data(wilson_raw_data, max_expr = "low")
-cdata = tmp_data[[1]] #continuous data
-ddata = tmp_data[[2]] #discretised dataOnce data are loaded and preprocessed, filter the cell types or genes to be included in the analysis if needed. It is advisable to reduce the number of genes to be included if the computation takes too long to complete. In this example, genes are not filtered as all genes that are present in both expression data and Boolean model are used automatically.
-# (3) Filter cell types.
-cell_ind = grepl("cmp", rownames(cdata)) | grepl("gmp", rownames(cdata)) | grepl("mep",
- rownames(cdata))
-fcdata = cdata[cell_ind, ] #select only relevant cells.
-fddata = ddata[cell_ind, ]6.3.4 Add extra genes to the initial Boolean model
-Extra genes can be added to the initial model using grow_bmodel. The function will add extra genes into the initial model with empty update functions.
# (4) Adding extra genes to the initial Boolean model. extra_genes =
-# setdiff(colnames(wilson_raw_data), bmodel@target) print(extra_genes) #to
-# view available genes to be added.
-add_gene = "ldb1" #genes to be added: ldb1
-grown_bmodel = grow_bmodel(add_gene, bmodel)6.3.5 Estimate initial state for the extra genes
-Initial state needs to be modify to include the initial expression of the extra genes. The initial state of the extra genes can be set manually, or it can be estimated from the data if the data contain multiple cell types with known relationships. In this example, CMPs are known to be at developmental upstream of erythro-myeloid differentiation, therefore initial state can be estimated by taking the average expression of the extra genes in CMPs.
-# (5) Estimating initial state for the extra genes. (estimating from CMPs)
-tmp_istate = mean(cdata[grepl("cmp", rownames(cdata)), add_gene])
-tmp_istate = matrix(round(tmp_istate), nrow = 1)
-colnames(tmp_istate) = add_gene
-grown_istate = cbind(istate, tmp_istate)
-grown_istate = initialise_data(grown_istate)6.3.6 Run model training
-To reconstruct a Boolean model from an expression data, run model_train.
In this example, model_train takes a few minutes to be completed on a single core. If this steps take a very long time to complete, do consider using the parallel processing option as described above.
You will receive a BoolModel object at the end of the model training process. The BoolModel object can be visualise using plotBM, which is based on igraph package. For easier manipulation, output the Boolean model using outgraph_model and display it with Cytoscape or Gephi.
Note that this example takes a long time to run. The use of parallel processing is recommended.
-# (6) Inferring Boolean model.
-final_model = model_train(cdata = fcdata, ddata = fddata, bmodel = grown_bmodel,
- istate = grown_istate, verbose = T)
-
-# (7) Visualise the Boolean model generated.
-plotBM(final_model)