empirical_controls: Empirical controls calculation function

In leekgroup/sig2trial: Pipeline R package for creating simple gene signatures and testing their value in a clinical trial setting

Description

Generate candidate top-scoring pairs (TSPs) by finding "empirical control" genes within quantiles of the data.

Usage

empirical_controls(exprs, n = 40, q = 4)

Arguments

exprs	An p x m input matrix of gene expression values with p genes and m samples. This matrix MUST have rownames.
n	Number of pairs to create within each quantile bin (default 40)
q	Number of quantiles desired (default 4)

Details

We intend to use TSPs as the features in our gene signature. Examining all features would require generating $p$ choose 2 comparisons, most of which would be uninformative. We prioritize the creation of features where one gene is expressed at a constant level across classes and the other gene is expressed high and low across classes. To unearth pairs that behave like this, we rank all genes by their average expression and break them into quantile groups. We then rank all genes within each group by their variance and compute comparisons between the top $n$ most variable and least variable genes. The result of this function provides a matrix of size $(n*q)$ x $m$, where each row is a binary vector indicating the value of I($g_1 > g_2$) for two selected.

Value

A (n*q) x m matrix with TSPs in rows and samples in columns

leekgroup/sig2trial documentation built on May 20, 2019, 11:31 p.m.