DoModelSetup: This function performs the following tasks: 1. Run IntLIM to...

In ncats/MultiOmicsGraphPrediction: Network-Based Ensemble Prediction from IntLIM Predictors

This function performs the following tasks: 1. Run IntLIM to obtain all pairwise models. 2. Filter the IntLIM results to obtain a subset of models. 3. Predict the phenotype of all training data given each pairwise model. 4. Build the co-regulation graph using the set of pairwise models. 5. Initialize model with parameters.

Description

This function performs the following tasks: 1. Run IntLIM to obtain all pairwise models. 2. Filter the IntLIM results to obtain a subset of models. 3. Predict the phenotype of all training data given each pairwise model. 4. Build the co-regulation graph using the set of pairwise models. 5. Initialize model with parameters.

Usage

DoModelSetup(
  inputData,
  stype,
  outcomeType = 1,
  independentVarType = 2,
  continuous = TRUE,
  pvalcutoff = 1,
  rsquaredCutoff = 0,
  coeffPercentile = 0,
  metaFeatureList = c("pdf", "interactionpval", "interactioncoef", "analytecoef",
    "localerr"),
  k = 2,
  eigStep = 1,
  colIdInd = "databaseId",
  colIdOut = "databaseId",
  edgeTypeList = c("shared.outcome.analyte", "shared.independent.analyte"),
  learningRate = 0.2,
  covar = c(),
  maxIterations = 1000,
  convergenceCutoff = 0.001,
  optimizationType = "SGD",
  initialMetaFeatureWeights = 0,
  corrCutoff = 0.7
)

Arguments

inputData	An object with the following fields:
stype	The phenotype (outcome) to predict. This can be either a categorical or numeric outcome.
outcomeType	The outcome type (1 or 2)
independentVarType	The independent variable type (1 or 2)
continuous	Whether or not the outcome is continuous. Default is TRUE.
pvalcutoff	Predictors with p-values higher than this cutoff will be discarded.
rsquaredCutoff	Predictors with R^2 values below this cutoff will be discarded.
coeffPercentile	Predictors with interaction coefficient percentiles below this cutoff will be discarded.
metaFeatureList	A list of the valid metrics to include. Valid metrics are "pdf", "localerr", "globalerr", and "pathway".
k	The number of nearest neighbors to consider in localerr.
eigStep	The number of eigenvectors to step by during the evaluation in localerr. Note that this must be less than the number of samples in localerr. Default = 10.
colIdInd	The ID of the column that has the analyte ID's for the independent variable. If blank, then the existing ID's are used.
colIdOut	The ID of the column that has the analyte ID's for the outcome variable. If blank, then the existing ID's are used.
edgeTypeList	List containing one or more of the following to include in the line graph: - "shared.outcome.analyte" - "shared.independent.analyte" - "analyte.chain"
learningRate	Learning rate to use during training. Default is 0.2
covar	The clinical covariates to include in the model. These should be the same covariates that were included when running the IntLIM linear models.
maxIterations	Maximum number of iterations. Default is 1,000.
convergenceCutoff	Cutoff for convergence. Default is 0.001.
optimizationType	Type of optimization. May be "SGD", "momentum", "adagrad", or "adam". Default is "SGD".
initialMetaFeatureWeights	Initial weights for model meta-features. Default is 0, which results in each meta-feature being given equal weight.
corrCutoff	Predictors with correlations higher than this cutoff will be clustered together, and a representative predictor will be chosen.

ncats/MultiOmicsGraphPrediction documentation built on Aug. 23, 2023, 9:19 a.m.