bnem: Boolean Nested Effects Model main function

In MartinFXP/bnem: Training of logical models from indirect measurements of perturbation experiments

Boolean Nested Effects Model main function

Description

This function takes a prior network and normalized perturbation data as input and trains logical functions on that prior network

Usage

bnem(
  search = "greedy",
  fc = NULL,
  expression = NULL,
  egenes = NULL,
  pkn = NULL,
  design = NULL,
  stimuli = NULL,
  inhibitors = NULL,
  signals = NULL,
  CNOlist = NULL,
  model = NULL,
  sizeFac = 10^-10,
  NAFac = 1,
  parameters = list(cutOffs = c(0, 1, 0), scoring = c(0.1, 0.2, 0.9)),
  parallel = NULL,
  method = "cosine",
  relFit = FALSE,
  verbose = TRUE,
  reduce = TRUE,
  parallel2 = 1,
  initBstring = NULL,
  popSize = 100,
  pMutation = 0.5,
  maxTime = Inf,
  maxGens = Inf,
  stallGenMax = 10,
  relTol = 0.01,
  priorBitString = NULL,
  selPress = c(1.2, 1e-04),
  fit = "linear",
  targetBstring = "none",
  elitism = NULL,
  inversion = NULL,
  selection = c("t"),
  type = "SOCK",
  exhaustive = FALSE,
  delcyc = FALSE,
  seeds = 1,
  maxSteps = Inf,
  node = NULL,
  absorpII = TRUE,
  draw = TRUE,
  prior = NULL,
  maxInputsPerGate = 2
)

Arguments

search	Type of search heuristic. Either "greedy", "genetic" or "exhaustive". "greedy" uses a greedy algorithm to move through the local neighbourhood of a initial hyper-graph. "genetic" uses a genetic algorithm. "exhaustive" searches through the complete search space and is not recommended.
fc	m x l matrix of foldchanges of gene expression values or equivalent input (normalized pvalues, logodds, ...) for m E-genes and l contrasts. If left NULL, the gene expression data is used to calculate naive foldchanges.
expression	Optional normalized m x l matrix of gene expression data for m E-genes and l experiments.
egenes	list object; each list entry is named after an S-gene and contains the names of egenes which are potential children
pkn	Prior knowledge network as output by CellNOptR::readSIF.
design	Optional n x l design matrix with n S-genes and l experiments. If available. If kept NULL, bnem needs either stimuli, inhibitors or a CNOlist object.
stimuli	Character vector of stimuli names.
inhibitors	Character vector of inhibitors.
signals	Optional character vector of signals. Signals are S-genes, which can directly regulate E-genes. If left NULL, all stimuli and inhibitors are defined as signals.
CNOlist	CNOlist object (see package CellNOptR), if available.
model	Model object including the search space, if available. See CellNOptR::preprocessing.
sizeFac	Size factor penelizing the hyper-graph size.
NAFac	factor penelizing NAs in the data.
parameters	parameters for discrete case (not recommended); has to be a list with entries cutOffs and scoring: cutOffs = c(a,b,c) with a (cutoff for real zeros), b (cutoff for real effects), c = -1 for normal scoring, c between 0 and 1 for keeping only relevant between -1 and 0 for keeping only a specific quantile of E-genes, and c > 1 for keeping the top c E-genes; scoring = c(a,b,c) with a (weight for real effects), c (weight for real zeros), b (multiplicator for effects/zeros between a and c);
parallel	Parallelize the search. An integer value specifies the number of threads on the local machine or a list object as in list(c(1,2,3), c("machine1", "machine2", "machine3")) specifies the threads distributed on different machines (local or others).
method	Scoring method can be "cosine", a correlation, or a distance measure. See ?cor and ?dist for details.
relFit	if TRUE a relative fit for each E-gene is computed (not recommended)
verbose	TRUE for verbose output
reduce	if TRUE reduces the search space for exhaustive search
parallel2	if TRUE parallelises the starts and not the search itself
initBstring	Binary vector for the initial hyper-graph.
popSize	Population size (only "genetic").
pMutation	Probability between 0 and 1 for mutation (only "genetic").
maxTime	Define a maximal time (seconds) for the search.
maxGens	Maximal number of generations (only "genetic").
stallGenMax	Maximum number of stall generations (only "genetic").
relTol	Score tolerance for networks defined as optimal but with a lower score as the real optimum (only "genetic").
priorBitString	Binary vector defining hyper-edges which are added to every hyper-graph. E.g. if you know hyper-edge 55 is definitly there and to fix that, set priorBitString[55] <- 1 (only "genetic").
selPress	Selection pressure between 1 and 2 (if fit="linear") and greater 2 (for fit "nonlinear") for the stochastic universal sampling (only "genetic").
fit	"linear" or "nonlinear fit for stochastic universal sampling
targetBstring	define a binary vector representing a network; if this network is found, the computation stops
elitism	Number of best hyper-graphs transferred to the next generation (only "genetic").
inversion	Number of worst hyper-graphs for which their binary strings are inversed (only "genetic").
selection	"t" for tournament selection and "s" for stochastic universal sampling (only "genetic").
type	type of the paralellisation on multpile machines (default: "SOCK")
exhaustive	If TRUE an exhaustive search is conducted if the genetic algorithm would take longer (only "genetic").
delcyc	If TRUE deletes cycles in all hyper-graphs (not recommended).
seeds	how many starts for the greedy search? (default: 1); uses the n-dimensional cube (n = number of S-genes) to maximize search space coverage
maxSteps	Maximal number of steps (only "greedy").
node	vector of S-gene names, which are used in the greedy search; if node = NULL all nodes are considered
absorpII	Use inverse absorption (default: TRUE).
draw	If TRUE draws the network evolution.
prior	Binary vector. A 1 specifies hyper-edges which should not be optimized (only "greedy").
maxInputsPerGate	If no model is supplied, one is created with maxInputsPerGate as maximum number of parents for each hyper-edge.

Value

List object including the optimized hyper-graph, its corresponding binary vector for full hyper-graph and optimized scores.

Author(s)

Martin Pirkl

See Also

nem

Examples

sifMatrix <- rbind(c("A", 1, "B"), c("A", 1, "C"), c("B", 1, "D"),
c("C", 1, "D"))
temp.file <- tempfile(pattern="interaction",fileext=".sif")
write.table(sifMatrix, file = temp.file, sep = "\t",
row.names = FALSE, col.names = FALSE,
quote = FALSE)
PKN <- CellNOptR::readSIF(temp.file)
CNOlist <- dummyCNOlist("A", c("B","C","D"), maxStim = 1,
maxInhibit = 2, signals = c("A", "B","C","D"))
model <- CellNOptR::preprocessing(CNOlist, PKN, maxInputsPerGate = 100)
expression <- matrix(rnorm(nrow(slot(CNOlist, "cues"))*10), 10,
nrow(slot(CNOlist, "cues")))
fc <- computeFc(CNOlist, expression)
initBstring <- rep(0, length(model$reacID))
res <- bnem(search = "greedy", model = model, CNOlist = CNOlist,
fc = fc, pkn = PKN, stimuli = "A", inhibitors = c("B","C","D"),
parallel = NULL, initBstring = initBstring, draw = FALSE, verbose = FALSE,
maxSteps = Inf)

MartinFXP/bnem documentation built on Nov. 5, 2024, 11:57 a.m.