In vjcitn/drosmap: fruitfly cell fate map from NMF on spatial expression matrix

library(knitcitations)
library(bibtex)
allbib = read.bibtex("sigiwu.bib")

Introduction

The recent PNAS paper of Wu and colleagues (r citet(allbib[["Wu2016"]])) demonstrates the use of cluster analysis of spatially organized data to identify gene expression signatures for regions of the cell-fate map of D. melanogaster. We have repackaged some of their data and code to demonstrate the key ideas.

Views of the cell-fate map

For developmental stages 4-6 of Drosophila (1.3-2h after egg deposit at 25C) a classic cell-fate map has been published (r citet(allbib[["Hartenstein1985"]])).

library(png)
library(grid)
im = readPNG("map1.png")
grid.raster(im)

Berkeley Drosophila Genome Project expression data

The image for "opa" in developmental stages 4-5 (1.3-3.0h after egg deposit)

im = readPNG("bdgpOPA.png")
grid.raster(im)

Setup for digitized expression patterns published by Wu et al. PNAS 2016

library(drosmap)
data(expressionPatterns)
data(uniqueGenes)
dim(expressionPatterns[,uniqueGenes])
data(template)
dim(template)
args(imageBatchDisplay)

Wu et al. reregistered, normalized colormap for opa

imageBatchDisplay(expressionPatterns[, "opa", drop=FALSE], 
  nrow=1, ncol=1, template=template[,-1])

Wu et al. reregistered, normalized colormaps for 25 genes (of 701 with 'spatially restricted expression')

imageBatchDisplay(expressionPatterns[, uniqueGenes[1:25]],
  nrow=5, ncol=5, template=template[,-1])

The NMF analysis of the 405 x 701 expression matrix

The expression patterns in the matrix $X$ with rows corresponding to positions in the blastocyst ellipse and column corresponding to genes are re-expressed via $X \approx DA$, where all entries in $D$ and $A$ are nonnegative. Matrix $D$ is referred to as basis, and matrix $A$ holds 'mixture coefficients'. We'll accept the statement that a rank 21 basis is adequate.

library(NMF)
uex = expressionPatterns[,uniqueGenes]
mm = nmf(data.matrix(uex), rank=21)
mm

The rows of the basis matrix can be clustered to exhibit its structure.

basismap(mm)

These can be projected back into the blastocyst image space. The 21 principal patterns are then

imageBatchDisplay(basis(mm), nrow=5, ncol=5, template=template[,-1])

References

bibliography() #style="markdown")

vjcitn/drosmap documentation built on May 3, 2019, 6:14 p.m.