Data Analysis Modules: panoply_association

panoply_association

Description

Performs association analysis to identify differential markers for classes of interest using a moderated t-test (for binary classes) or F-test (for categorical multi-level classes). The significant markers are then ranked using a combination of p-values and variable importance in accurate classifiers.

Classes used for the analysis are derived from the annotations provided in the groups file, as long as each level in a class has at least 3 samples.

This module performs the following steps:

• Filters out (removes) features with >50% missing values, and imputes missing values using k-NN imputation.
• Runs marker selection using LIMMA.
• Uses statistically significant markers to build classifiers using PLS, PAM, GLMNET, and RF models.
• Extracts marker importance (a measure of how crucial the marker was for training an accurate classifier) for input markers and combines all the ranks into a global ranking using rank aggregation.
• If provided with test data, runs prediction using all classifiers.
• Plots a heatmap of training data showing significant markers.
• Runs GSEA for (binary) classes with only two levels.

Input

Required inputs:

• inputData: (.tar file) tarball from panoply_parse_sm_table or other PANOPLY module;
  (.gct file) normalized/filtered input if standalone is TRUE
• type: (String) (proteome) data type
• standalone: (String) set to TRUE to run as a self-contained module;
  if TRUE the analysisDir and groupsFile inputs are required
• yaml: (.yaml file) parameters in yaml format

Optional inputs:

• analysisDir: (String) name of analysis directory
• groupsFile: (.csv file) subset of sample annotations, providing classes for association analysis
• fdr_assoc: (Float, default = 0.01) FDR cutoff value for significance
• sample_na_max: (Float, default = 0.8) maximum allowed fraction of NA values per sample/column; error if violated.
• nmiss_factor: (Float, default = 0.5) features (genes, proteins, PTM sites) with more than nmiss_factor fraction of NA values will be removed from the analysis
• duplicate_gene_policy: (String, default = 'maxvar') method used to combine duplicate genes (when mapping protein accession or PTM site to gene symbols) for running GSEA; possible options are:
  • maxvar: select row with largest variance
  • union: union of binary (0/1) values in all rows (e.g for mutation status)
  • median: median of values in all rows (for each column/sample)
  • mean: mean of values in all rows (for each column/sample)
  • min: minimum of values in all rows (for each column/sample)
• gene_id_col: (String, default = 'geneSymbol') name of sample annotation column containing gene ids.
• outFile: (String, default = "panoply_association-output.tar") output .tar file name

Output

• outputs Tarball of files containing the following in the association subdirectory, for each class vector considered for association analysis:
  • List of significant differential markers derived using LIMMA (*-markers-all-fdr*.csv) and p-values for all input features (*-markers-all*.csv)
  • Marker importance for significant markers, along with final rank (*-markerimp-fdr*.csv)
  • Heatmap of significant differential markers (*-markers-heatmap.pdf)
  • Classifier performance contingency tables (*-analysis-model-results.txt)
  • Table of prediction results for training data (*-train-results-*.csv) and testing data (*-test-results.csv) using all classifiers
  • GSEA outputs, along with .gct. and .cls input files, for binary classes (in *-gsea-analysis/ subdirectory).