While single cell RNA-seq (scRNA-seq) gives us insights on biological
systems with unprecedented resolution, as tissue dissociation is
required for scRNA-seq, spatial context of gene expression is lost.
novoSpaRc is a way to reconstruct the spatial context of gene
expression by optimal transport. novoSpaRc is described in the paper
Charting a Tissue from Single Cell
Transcriptomes. The
paper authors implemented this method in Python, which can be found
here. This package is an R
implementation of this method.
In short, what novoSpaRc does is to find a probabilistic assignment of
cells to locations by aligning structural similarities between the
graphs generated for single cells in expression space and physical
space, with or without taking into account an in situ atlas that
quantifies expression of some landmark genes in spatial locations.
This package is not yet on CRAN or Bioconductor. Please install it with
devtools::install_github("lambdamoses/novoSpaRc")We also strongly recommend an optimized BLAS, as this package makes heavy use of matrix multiplications, and the matrices can be large for larger datasets. The default BLAS that comes with R is not optimized; an optimized BLAS can speed up matrix multiplications several times. To use an optimized BLAS from R, you can install Microsoft R Open, which uses the optimized Intel Math Kernel Library (MKL) on Windows and Linux and Acceleration Framework on MacOS for BLAS. Other optimized BLAS are OpenBLAS and ATLAS.
For MacOS users, you may change the version of BLAS used by R without
reinstalling R by adding a symbolic link from libRblas.dylib to the
desired BLAS. For example, for R installed from CRAN binary, here is how
to make R use the BLAS from Acceleration Framework, which comes with
MacOS:
cd /Library/Frameworks/R.framework/Resources/lib
ln -sf /System/Library/Frameworks/Accelerate.framework/Frameworks/vecLib.framework/libBLAS.dylib \
libRblas.dylibFor changing BLAS on Windows and Ubuntu, see this blog post. For Fedora, see this post.