Symphony

Efficient and precise single-cell reference atlas mapping with Symphony

Kang et al. (Nat Comm, 2021)

For Python users, check out the symphonypy package by Kseniya Petrova and Sergey Isaev.

Installation

Symphony is available on CRAN:

install.packages("symphony")

Install the development version of Symphony from GitHub use:

# install.packages("devtools")
devtools::install_github("immunogenomics/symphony")

Install should take \<10 mins (pending no major issues). See installation notes below.

Usage/Demos

Tutorials

Check out the quick start (\<5 min) PBMCs tutorial for an example of how to build a custom reference and map to it.
Check out the pre-built references tutorial for examples of how to map to provided Symphony references pre-built from the datasets featured in the manuscript.

Downloading pre-built references:

You can download pre-built references from Zenodo.

Reference building

Option 1: Starting from existing Harmony object

This function compresses an existing Harmony object into a Symphony reference that enables query mapping. We recommend this option for most users since it allows your code to be more modular and flexible.


# Run Harmony to integrate the reference cells
ref_harmObj = harmony::HarmonyMatrix(
        data_mat = t(Z_pca_ref),   # starting embedding (e.g. PCA, CCA) of cells
        meta_data = ref_metadata,  # dataframe with cell metadata
        theta = c(2),              # cluster diversity enforcement
        vars_use = c('donor'),     # variable to integrate out
        nclust = 100,              # number of clusters in Harmony model
        max.iter.harmony = 10,     # max iterations of Harmony
        return_object = TRUE,      # set to TRUE to return the full Harmony object
        do_pca = FALSE             # do not recompute PCs
)

# Build Symphony reference
reference = buildReferenceFromHarmonyObj(
        ref_harmObj,            # output object from HarmonyMatrix()
        ref_metadata,           # dataframe with cell metadata
        vargenes_means_sds,     # gene names, means, and std devs for scaling
        loadings,               # genes x PCs
        verbose = TRUE,         # display output?
        do_umap = TRUE,         # run UMAP and save UMAP model to file?
        save_uwot_path = '/absolute/path/uwot_model_1' # filepath to save UMAP model)

Note that vargenes_means_sds requires column names c('symbol', 'mean', 'stddev') (see tutorial example).

Option 2: Starting from reference genes by cells matrix

This function performs all steps of the reference building pipeline including variable gene selection, scaling, PCA, Harmony, and Symphony compression.

# Build reference
reference = symphony::buildReference(
    ref_exp,                   # reference expression (genes by cells)
    ref_metadata,              # reference metadata (cells x attributes)
    vars = c('donor'),         # variable(s) to integrate over
    K = 100,                   # number of Harmony soft clusters
    verbose = TRUE,            # display verbose output
    do_umap = TRUE,            # run UMAP and save UMAP model to file
    do_normalize = FALSE,      # perform log(CP10k) normalization on reference expression
    vargenes_method = 'vst',   # variable gene selection method: 'vst' or 'mvp'
    vargenes_groups = 'donor', # metadata column specifying groups for variable gene selection within each group
    topn = 2000,               # number of variable genes (per group)
    theta = 2,                 # Harmony parameter(s) for diversity term
    d = 20,                    # number of dimensions for PCA
    save_uwot_path = 'path/to/uwot_model_1', # file path to save uwot UMAP model
    additional_genes = NULL    # vector of any additional genes to force include
)

Query mapping

Once you have a prebuilt reference (e.g. loaded from a saved .rds R object), you can directly map cells from a new query dataset onto it starting from query gene expression.

# Map query
query = mapQuery(query_exp,             # query gene expression (genes x cells)
                 query_metadata,        # query metadata (cells x attributes)
                 reference,             # Symphony reference object
                 vars = NULL,           # Query batch variables to harmonize over (NULL treats query as one batch)
                 do_normalize = FALSE,  # perform log(CP10k) normalization on query (set to FALSE if already normalized)
                 do_umap = TRUE)        # project query cells into reference UMAP

query$Z contains the harmonized query feature embedding.

If your query itself has multiple sources of batch variation you would like to integrate over (e.g. technology, donors, species), you can specify them in the vars parameter: e.g. vars = c('donor', 'technology')

Installation notes

System requirements:

Symphony has been successfully installed on Linux and Mac OS X using the devtools package to install from GitHub.

Dependencies:

R>=3.6.x
RANN
data.table
irlba
stats
tibble
utils
uwot
Matrix
Rcpp
magrittr
methods
rlang
ggplot2
RColorBrewer
ggrastr
ggrepel

Troubleshooting:

You may need to install the latest version of devtools (because of the recent GitHub change from “master” to “main” terminology, which can cause install_github to fail).
You may also need to install the lastest version of Harmony:

devtools::install_github("immunogenomics/harmony")

We have been notified of the following installation errors regarding systemfonts, textshaping, and ragg (which are all required by ggrastr):

# error when installing systemfonts
ft_cache.h:9:10: fatal error: ft2build.h: No such file or directory

# error when installing textshaping
Configuration failed to find the harfbuzz freetype2 fribidi library

# error when installing ragg
<stdin>:1:10: fatal error: ft2build.h: No such file or directory

These errors are not inherent to the Symphony package and we cannot fix them directly. However, as a workaround, you can install systemfonts, textshaping, and ragg separately using install.packages() and specify the path to the required files (replacing /path/to below with the path to the appropriate include directory containing the files).

# fix to install systemfonts
withr::with_makevars(c(CPPFLAGS="-I/path/to/include/freetype2/"), install.packages("systemfonts"))

# fix to install textshaping
withr::with_makevars(c(CPPFLAGS="-I/path/to/include/harfbuzz/ -I/path/to/include/fribidi/ -I/path/to/include/freetype2/"), install.packages("textshaping"))

# fix to install ragg
withr::with_makevars(c(CPPFLAGS="-I/path/to/include/freetype2/"), install.packages("ragg"))

Reproducing results from manuscript

Code to reproduce Symphony results from the Kang et al. manuscript is available on github.com/immunogenomics/symphony_reproducibility.

immunogenomics / symphony

readme