Inferelator: An R repository from ChristophH

Call the inferelator script from the base directory (the one containing this 
README) with a job config file as argument. 

Example call: Rscript inferelator.R jobs/dream4_cfg.R



--------------------------------------------------------------------------------
Default parameters and a brief explanation of each one
--------------------------------------------------------------------------------

PARS$input.dir <- 'input/dream4'  # path to the input files

PARS$exp.mat.file <- 'expression.tsv'  # required; see definition below
PARS$tf.names.file <- 'tf_names.tsv'  # required; see definition below
PARS$meta.data.file <- 'meta_data.tsv'  # assume all steady state if NULL
PARS$priors.file <- 'gold_standard.tsv'  # no priors if NULL
PARS$gold.standard.file <- 'gold_standard.tsv'  # no evaluation if NULL
PARS$leave.out.file <- NULL  # file with list of conditions that will be ignored
PARS$randomize.expression <- FALSE  # whether to scramble input expression

PARS$job.seed <- 42  # random seed; can be NULL
PARS$save.to.dir <- file.path(PARS$input.dir, date.time.str)  # output directory
PARS$num.boots <- 20  # number of bootstraps; no bootstrapping with a value of 1
PARS$max.preds <- 10  # max number of predictors based on CLR to pass to model
                      # selection method
PARS$mi.bins <- 10  # number of bins to use for mutual information calculation
PARS$cores <- 8  # number of cpu cores

PARS$delT.max <- 110  # max number of time units allowed between time series 
                      # conditions
PARS$delT.min <- 0  # min number of time units allowed between time series 
                    # conditions
PARS$tau <- 45  # constant related to half life of mRNA (see Core model)

PARS$perc.tp <- 0  # percent of true priors that will be used; can be vector
PARS$perm.tp <- 1  # number of permutations of true priors
PARS$perc.fp <- 0  # percent of false priors (100 = as many false priors as 
                   # there are true priors); can be vector
PARS$perm.fp <- 1  # number of permutations of false priors
PARS$pr.sel.mode <- 'random'  # prior selection mode: 'random' or 'tf'
                              # if 'random', the true priors are randomly chosen
                              # from all priors edges, if 'tf', 
                              # PARS$perc.tp is interpreted as the percent of
                              # TFs to use for true priors and all interactions
                              # for the chosen TFs will be used

PARS$eval.on.subset <- FALSE  # whether to evaluate only on the part of the 
                              # network that has connections in the gold
                              # standard; if TRUE false priors will only be 
                              # drawn from that part of the network

PARS$method <- 'BBSR'  # which method to use; either 'MEN' or 'BBSR'
PARS$prior.weight <- 1  # the weight for the priors; has to be larger than 1
                        # for priors to have an effect

PARS$use.tfa <- FALSE  # whether to estimate transcription factor activities and
                       # use those in the regression models
                       # if TRUE, interactions in priors file shoud be signed,
                       # i.e. -1 for repression and +1 for activation
PARS$prior.ss <- FALSE # whether to also sub-sample from the prior matrix during
                       # each bootstrap; if TRUE, priors are sampled randomly with 
                       # replacement; if FALSE, all priors are used as is

PARS$output.summary <- TRUE  # write a summary tsv and RData file of network

PARS$output.report <- TRUE  # create html network report

PARS$output.tf.plots <- TRUE  # create png files with plots of TFs and targets

--------------------------------------------------------------------------------
Required Input Files
--------------------------------------------------------------------------------

expression.tsv
--------------
expression values; must include row (genes) and column (conditions) names

tf_names.tsv
------------
one TF name on each line; must be subset of the row names of the expression data



--------------------------------------------------------------------------------
Optional Input Files
--------------------------------------------------------------------------------

meta_data.tsv
-------------
the meta data describing the conditions; must include column names;
has five columns:
isTs: TRUE if the condition is part of a time-series, FALSE else
is1stLast: "e" if not part of a time-series; "f" if first; "m" middle; "l" last
prevCol: name of the preceding condition in time-series; NA if "e" or "f"
del.t: time in minutes since prevCol; NA if "e" or "f"
condName: name of the condition

priors.tsv
----------
matrix of 0 and 1 indicating whether we have prior knowledge in 
the interaction of one TF and a gene; one row for each gene, one column for 
each TF; must include row (genes) and column (TF) names

gold_standard.tsv
-----------------
needed for validation; matrix of 0 and 1 indicating whether there is an 
interaction between one TF and a gene; one row for each gene, one column for 
each TF; must include row (genes) and column (TF) names



--------------------------------------------------------------------------------
Output Files
--------------------------------------------------------------------------------

One or more betas_frac_tp_X_perm_X--frac_fp_X_perm_X_X.RData files. One file
per true and false prior and prior weight combination. Each RData file contains
two lists of length PARS$num.boots where every entry is a matrix of betas and
confidence scores (rescaled betas) respectively.

One or more combinedconf_frac_tp_X_perm_X--frac_fp_X_perm_X_X.RData files with
one matrix each. The matrix is the rank-combined version of the confidence
scores of all bootstraps.

A params_and_input.RData file with data objects holding the user set parameters,
and input and input-derived objects.
ChristophH/Inferelator
