QSep-class: Quantify resolution of a spatial proteomics experiment

In lgatto/pRoloc: A unifying bioinformatics framework for spatial proteomics

Quantify resolution of a spatial proteomics experiment

Description

The QSep infrastructure provide a way to quantify the resolution of a spatial proteomics experiment, i.e. to quantify how well annotated sub-cellular clusters are separated from each other.

The QSep function calculates all between and within cluster average distances. These distances are then divided column-wise by the respective within cluster average distance. For example, for a dataset with only 2 spatial clusters, we would obtain

	c_1	c_2
c_1	d_11	d_12
c_2	d_21	d_22

Normalised distance represent the ratio of between to within average distances, i.e. how much bigger the average distance between cluster c_i and c_j is compared to the average distance within cluster c_i.

	c_1	c_2
c_1	1	\frac{d_12}{d_22}
c_2	\frac{d_21}{d_11}	1

Note that the normalised distance matrix is not symmetric anymore and the normalised distance ratios are proportional to the tightness of the reference cluster (along the columns).

Missing values only affect the fractions containing the NA when the distance is computed (see the example below) and further used when calculating mean distances. Few missing values are expected to have negligible effect, but data with a high proportion of missing data will will produce skewed distances. In QSep, we take a conservative approach, using the data as provided by the user, and expect that the data missingness is handled before proceeding with this or any other analysis.

Objects from the Class

Objects can be created by calls using the constructor QSep (see below).

Slots

x:

Object of class "matrix" containing the pairwise distance matrix, accessible with qseq(., norm = FALSE).

xnorm:

Object of class "matrix" containing the normalised pairwise distance matrix, accessible with qsep(., norm = TRUE) or qsep(.).

object:

Object of class "character" with the variable name of MSnSet object that was used to generate the QSep object.

.__classVersion__:

Object of class "Versions" storing the class version of the object.

Extends

Class "Versioned", directly.

Methods and functions

QSeq

signature(object = "MSnSet", fcol = "character"): constructor for QSep objects. The fcol argument defines the name of the feature variable that annotates the sub-cellular clusters. Non-marker proteins, that are marked as "unknown" are automatically removed prior to distance calculation.

qsep

signature{object = "QSep", norm = "logical"}: accessor for the normalised (when norm is TRUE, which is default) and raw (when norm is FALSE) pairwise distance matrices.

names

signature{object = "QSep"}: method to retrieve the names of the sub-celluar clusters originally defined in QSep's fcol argument. A replacement method names(.) <- is also available.

summary

signature(object = "QSep", ..., verbose = "logical"): Invisible return all between cluster average distances and prints (when verbose is TRUE, default) a summary of those.

levelPlot

signature(object = "QSep", norm = "logical", ...): plots an annotated heatmap of all normalised pairwise distances. norm (default is TRUE) defines whether normalised distances should be plotted. Additional arguments ... are passed to the levelplot.

plot

signature(object = "QSep", norm = "logical"...): produces a boxplot of all normalised pairwise distances. The red points represent the within average distance and black points between average distances. norm (default is TRUE) defines whether normalised distances should be plotted.

Author(s)

Laurent Gatto <[e-mail]>

References

Assessing sub-cellular resolution in spatial proteomics experiments Laurent Gatto, Lisa M Breckels, Kathryn S Lilley bioRxiv 377630; doi: https://doi.org/10.1101/377630

Examples

## Test data from Christoforou et al. 2016
library("pRolocdata")
data(hyperLOPIT2015)

## Create the object and get a summary
hlq <- QSep(hyperLOPIT2015)
hlq
summary(hlq)

## mean distance matrix
qsep(hlq, norm = FALSE)

## normalised average distance matrix
qsep(hlq)

## Update the organelle cluster names for better
## rendering on the plots
names(hlq) <- sub("/", "\n", names(hlq))
names(hlq) <- sub(" - ", "\n", names(hlq))
names(hlq)

## Heatmap of the normalised intensities
levelPlot(hlq)

## Boxplot of the normalised intensities
par(mar = c(3, 10, 2, 1))
plot(hlq)

## Boxplot of all between cluster average distances
x <- summary(hlq, verbose = FALSE)
boxplot(x)

## Missing data example, for 4 proteins and 3 fractions
x <- rbind(c(1.1, 1.2, 1.3), rep(1, 3), c(NA, 1, 1), c(1, 1, NA))
rownames(x) <- paste0("P", 1:4)
colnames(x) <- paste0("F", 1:3)

## P1 is the reference, against which we will calculate distances. P2
## has a complete profile, producing the *real* distance. P3 and P4 have
## missing values in the first and last fraction respectively.
x

## If we drop F1 in P3, which represents a small difference of 0.1, the
## distance only considers F2 and F3, and increases. If we drop F3 in
## P4, which represents a large distance of 0.3, the distance only
## considers F1 and F2, and decreases.  dist(x)

lgatto/pRoloc documentation built on Oct. 23, 2024, 12:51 a.m.