Package 'spqn'

Spatial quantile normalization

Description

The spqn package implements spatial quantile normalization (SpQN). This method was developed to remove a mean-correlation relationship in correlation matrices built from gene expression data. It can serve as pre-processing step prior to a co-expression analysis.

Details

See references for details on spatial quantile normalization.

The main function is normalize_correlation. We include a number of plotting functions for examining the mean-correlation relationship, see the vignette for examples.

References

Y Wang, SC Hicks, KD Hansen (2020). Co-expression analysis is biased by a mean-correlation relationship. bioRxiv 2020.02.13.944777. doi:10.1101/2020.02.13.944777

---

Spatial quantile normalization (SpQN)

Description

This method was developed to remove a mean-correlation relationship in correlation matrices built from gene expression data. It can serve as pre-processing step prior to a co-expression analysis.

Usage

normalize_correlation(cor_mat, ave_exp, ngrp, size_grp, ref_grp)

Arguments

cor_mat	A (square and symmetrix) correlation matrix.
ave_exp	A vector of expression levels, same length as the number of rows of the correlation matrix in cor_mat. For other types of data, ave_exp can be the vector corresponding to the row/column of the correlation matrix, whose dependency with the distribution of correlations need to be removed.
ngrp	Number of bins in each row/column to be used to partition the correlation matrix, integer.
size_grp	Size of the outer bins to be used to appriximate the distribution of the inner bins, in order to smooth the normalization. Note that the product of size_grp and ngrp must be equal or larger than than the row/column number of cor_mat, and there is no smoothness in the normalization when they are equal.
ref_grp	Location of the reference bin on the diagonal, whose distribution will be used as target distribution in the normalization, an integer.

Value

A normalized correlation matrix.

Examples

if(require(spqnData)){
  data(gtex.4k)
  cor_ori <- cor(t(assay(gtex.4k)))
  ave_logrpkm <- rowData(gtex.4k)$ave_logrpkm
  normalize_correlation(cor_ori, ave_exp = ave_logrpkm,
                        ngrp=10, size_grp=15, ref_grp=9)}

---

Get and plot the IQRs of submatrices of the correlation matrix.

Description

The get_IQR_condition_exp function computes the IQRs of a set of 10 by 10 same-size bins that partition the correlation matrix, ordered according to expression level.

The plot_IQR_condition_exp function plots the IQR for each bin among a set of 10 by 10 same-size bins that partition the correlation matrix, with IQR denoted by the width of boxes in the plot.

Usage

get_IQR_condition_exp(cor_mat, ave_exp)
plot_IQR_condition_exp(IQR_list)

Arguments

cor_mat	correlation matrix, generated by gene expression matrix, with genes sorted by average expression levels.
ave_exp	vector, average expression level of each gene for the normalized gene expression matrix.
IQR_list	List, output of get_IQR_condition_exp.

Value

A plot with boxes that shows the IQR of each bin

Note

The mnemonic for condition_exp is ‘conditional on expression’.

Examples

if(require(spqnData)) {
    data(gtex.4k)
    cor_mat <- cor(t(assay(gtex.4k)))
    ave_logrpkm <- rowData(gtex.4k)$ave_logrpkm
    IQR_list <- get_IQR_condition_exp(cor_mat, ave_exp = ave_logrpkm)
    plot_IQR_condition_exp(IQR_list)
    }

---

Plot the signal and background distribution of a correlation matrix.

Description

This function allows users to visualize the distributions of (assumed) signal and background, conditional on expression levels. The predicted signals are defined by the 0.1% highest correlations in each bin.

Usage

plot_signal_condition_exp(cor_mat, ave_exp, signal)

Arguments

cor_mat	Matrix, correlation matrix, generated by gene expression matrix
ave_exp	Vector, average expression level of each gene for the normalized expression matrix
signal	a value between 0 and 1 giving the fraction of correlations which should be considered signal. We often use a value of 0.001.

Value

Invoked for the side effect of producing a plot.

Note

The mnemonic for condition_exp is ‘conditional on expression’.

Examples

if(require(spqnData)) {
  data(gtex.4k)
  cor_mat <- cor(t(assay(gtex.4k)))
  ave_logrpkm <- rowData(gtex.4k)$ave_logrpkm
  plot_signal_condition_exp(cor_mat, ave_exp=ave_logrpkm, signal=0.05)}

---

Q-Q plot for examing the distributions across submatrices of a correlation matrix.

Description

We partition the correlation matrix into 10x10 bins of equal size, with genes ordered according to expression level. As reference bin, we choose the (9,9) bin (ie. the almost-highest expressed genes). We then make a QQ-plot of the (i,j)'th submatrix vs. the (9,9) submatrix. See the SpQN paper for detail on these choices.

Usage

qqplot_condition_exp(cor_mat,ave_exp, i,j)

Arguments

cor_mat	Matrix, correlation matrix, generated by gene expression matrix.
ave_exp	Vector, average expression level of each gene for the normalized expression matrix.
i	Integer, row number of the submatrix (see details).
j	Integer, column number of the submatrix (see details).

Value

Invoked for the side effect of producing a plot.

Note

The mnemonic for condition_exp is ‘conditional on expression’.

Examples

if(require(spqnData)) {
  data(gtex.4k)
  cor_mat <- cor(t(assay(gtex.4k)))
  ave_logrpkm <- rowData(gtex.4k)$ave_logrpkm
  qqplot_condition_exp(cor_mat, ave_exp=ave_logrpkm, 1, 1)
}

---