Cell omics such as single-cell genomics, proteomics, and microbiomics allow the characterisation of tissue and microbial community composition, which can be compared between conditions to identify biological drivers. This strategy has been critical to unveiling markers of disease progression, such as cancer and pathogen infection.
For cell omic data, no method for differential variability analysis exists, and methods for differential composition analysis only take a few fundamental data properties into account. Here we introduce sccomp, a generalised method for differential composition and variability analyses capable of jointly modelling data count distribution, compositionality, group-specific variability, and proportion mean-variability association, with awareness against outliers.
Sccomp is an extensive analysis framework that allows realistic data simulation and cross-study knowledge transfer. We demonstrate that mean-variability association is ubiquitous across technologies, highlighting the inadequacy of the very popular Dirichlet-multinomial modelling and providing essential principles for differential variability analysis.
We show that sccomp accurately fits experimental data, with a 50% incremental improvement over state-of-the-art algorithms. Using sccomp, we identified novel differential constraints and composition in the microenvironment of primary breast cancer.
sccomp
tests differences in cell type proportions from single-cell
data. It is robust against outliers, it models continuous and discrete
factors, and capable of random-effect/intercept modelling.
Please cite PNAS - sccomp: Robust differential composition and variability analysis for single-cell data
- Complex linear models with continuous and categorical covariates
- Multilevel modelling, with population fixed and random effects/intercept
- Modelling data from counts
- Testing differences in cell-type proportionality
- Testing differences in cell-type specific variability
- Cell-type information share for variability adaptive shrinkage
- Testing differential variability
- Probabilistic outlier identification
- Cross-dataset learning (hyperpriors).
sccomp
is based on cmdstanr
which provides the latest version of
cmdstan
the Bayesian modelling tool. cmdstanr
is not on CRAN, so we
need to have 3 simple step process (that will be prompted to the user is
forgot).
- R installation of
sccomp
- R installation of
cmdstanr
cmdstanr
call tocmdstan
installation
Bioconductor
if (!requireNamespace("BiocManager")) install.packages("BiocManager")
# Step 1
BiocManager::install("sccomp")
# Step 2
install.packages("cmdstanr", repos = c("https://stan-dev.r-universe.dev/", getOption("repos")))
# Step 3
cmdstanr::check_cmdstan_toolchain(fix = TRUE) # Just checking system setting
cmdstanr::install_cmdstan()
Github
# Step 1
devtools::install_github("stemangiola/sccomp")
# Step 2
install.packages("cmdstanr", repos = c("https://stan-dev.r-universe.dev/", getOption("repos")))
# Step 3
cmdstanr::check_cmdstan_toolchain(fix = TRUE) # Just checking system setting
cmdstanr::install_cmdstan()
Function | Description |
---|---|
sccomp_estimate |
Fit the model onto the data, and estimate the coefficients |
sccomp_remove_outliers |
Identify outliers probabilistically based on the model fit, and exclude them from the estimation |
sccomp_test |
Calculate the probability that the coefficients are outside the H0 interval (i.e. test_composition_above_logit_fold_change) |
sccomp_replicate |
Simulate data from the model, or part of the model |
sccomp_predict |
Predicts proportions, based on the model, or part of the model |
sccomp_remove_unwanted_variation |
Removes the variability for unwanted factors |
plot |
Plots summary plots to asses significance |
sccomp
can model changes in composition and variability. By default,
the formula for variability is either ~1
, which assumes that the
cell-group variability is independent of any covariate or
~ factor_of_interest
, which assumes that the model is dependent on the
factor of interest only. The variability model must be a subset of the
model for composition.
Of the output table, the estimate columns start with the prefix c_
indicate composition
, or with v_
indicate variability
(when
formula_variability is set).
sccomp_result =
single_cell_object |>
sccomp_estimate(
formula_composition = ~ type,
.sample = sample,
.cell_group = cell_group,
bimodal_mean_variability_association = TRUE,
cores = 1
) |>
sccomp_remove_outliers(cores = 1) |> # Optional
sccomp_test()
sccomp_result =
counts_obj |>
sccomp_estimate(
formula_composition = ~ type,
.sample = sample,
.cell_group = cell_group,
.count = count,
bimodal_mean_variability_association = TRUE,
cores = 1, verbose = FALSE
) |>
sccomp_remove_outliers(cores = 1, verbose = FALSE) |> # Optional
sccomp_test()
## Running standalone generated quantities after 1 MCMC chain, with 1 thread(s) per chain...
##
## Chain 1 finished in 0.0 seconds.
## Running standalone generated quantities after 1 MCMC chain, with 1 thread(s) per chain...
##
## Chain 1 finished in 0.0 seconds.
Here you see the results of the fit, the effects of the factor on composition and variability. You also can see the uncertainty around those effects.
sccomp_result
## # A tibble: 72 × 14
## cell_group parameter factor c_lower c_effect c_upper c_pH0 c_FDR v_lower
## <chr> <chr> <chr> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl>
## 1 B1 (Intercep… <NA> 0.886 1.05 1.23 0 0 -6.13
## 2 B1 typecancer type -1.16 -0.884 -0.623 0 0 NA
## 3 B2 (Intercep… <NA> 0.422 0.702 0.969 0 0 -5.78
## 4 B2 typecancer type -1.18 -0.810 -0.429 2.50e-4 3.12e-5 NA
## 5 B3 (Intercep… <NA> -0.638 -0.377 -0.130 1.55e-2 1.06e-3 -6.78
## 6 B3 typecancer type -0.606 -0.248 0.134 2.17e-1 4.81e-2 NA
## 7 BM (Intercep… <NA> -1.27 -1.01 -0.744 0 0 -7.36
## 8 BM typecancer type -0.715 -0.350 0.0280 9.88e-2 1.87e-2 NA
## 9 CD4 1 (Intercep… <NA> 0.146 0.318 0.503 8.25e-3 4.35e-4 -6.59
## 10 CD4 1 typecancer type -0.113 0.132 0.376 3.99e-1 1.26e-1 NA
## # ℹ 62 more rows
## # ℹ 5 more variables: v_effect <dbl>, v_upper <dbl>, v_pH0 <dbl>, v_FDR <dbl>,
## # count_data <list>
The estimated effects are expressed in the unconstrained space of the parameters. Similarly, to differential expression analysis that express change in terms of log fold change. However, for differences, in proportion, logit foold change must be used. This measure is harder to interpret and understand.
Therefore, we provide a more intuitive proportion, full change, that can be easier understood. However, these cannot be used to infer significance (use sccomp_test() instead), and a lot of care must be taken given the nonlinearity of these measure (1 fold increase from 0.0001 to 0.0002 carried a different weight that 1 fold increase from 0.4 to 0.8).
From your estimates, you can state which effects you are interested about (this can be a part of the full model, in case you want to not consider unwanted effects), and the two points you would like to compare.
In case of a chategorical variable, the starting and ending points are categories.
sccomp_result |>
sccomp_proportional_fold_change(
formula_composition = ~ type,
from = "healthy",
to = "cancer"
) |>
select(cell_group, statement)
## Loading model from cache...
## Running standalone generated quantities after 1 MCMC chain, with 1 thread(s) per chain...
##
## Chain 1 finished in 0.0 seconds.
## # A tibble: 36 × 2
## cell_group statement
## <chr> <glue>
## 1 B1 2.4-fold decrease (from 0.0528 to 0.0225)
## 2 B2 2.2-fold decrease (from 0.0373 to 0.0171)
## 3 B3 1.2-fold decrease (from 0.0126 to 0.0103)
## 4 BM 1.4-fold decrease (from 0.0068 to 0.005)
## 5 CD4 1 1.2-fold increase (from 0.0253 to 0.0298)
## 6 CD4 2 1.7-fold increase (from 0.0476 to 0.0827)
## 7 CD4 3 3.3-fold decrease (from 0.1019 to 0.0307)
## 8 CD4 4 1.2-fold increase (from 0.0016 to 0.0019)
## 9 CD4 5 1-fold increase (from 0.0302 to 0.0312)
## 10 CD8 1 1.2-fold increase (from 0.1027 to 0.1269)
## # ℹ 26 more rows
A plot of group proportion, faceted by groups. The blue boxplots
represent the posterior predictive check. If the model is likely to be
descriptively adequate to the data, the blue box plot should roughly
overlay with the black box plot, which represents the observed data. The
outliers are coloured in red. A box plot will be returned for every
(discrete) covariate present in formula_composition
. The colour coding
represents the significant associations for composition and/or
variability.
sccomp_result |>
sccomp_boxplot(factor = "type")
## Loading model from cache...
## Running standalone generated quantities after 1 MCMC chain, with 1 thread(s) per chain...
##
## Chain 1 finished in 0.0 seconds.
## Joining with `by = join_by(cell_group, sample)`
## Joining with `by = join_by(cell_group, type)`
A plot of estimates of differential composition (c_) on the x-axis and differential variability (v_) on the y-axis. The error bars represent 95% credible intervals. The dashed lines represent the minimal effect that the hypothesis test is based on. An effect is labelled as significant if bigger than the minimal effect according to the 95% credible interval. Facets represent the covariates in the model.
sccomp_result |>
plot_1D_intervals()
We can plot the relationship between abundance and variability. As we can see below, they are positively correlated, you also appreciate that this relationship is by model for single cell RNA sequencing data.
sccomp
models, these relationship to obtain a shrinkage effect on the
estimates of both the abundance and the variability. This shrinkage is
adaptive as it is modelled jointly, thanks for Bayesian inference.
sccomp_result |>
plot_2D_intervals()
You can produce the series of plots calling the plot
method.
sccomp_result |> plot()
seurat_obj |>
sccomp_estimate(
formula_composition = ~ 0 + type,
.sample = sample,
.cell_group = cell_group,
bimodal_mean_variability_association = TRUE,
cores = 1, verbose = FALSE
) |>
sccomp_test( contrasts = c("typecancer - typehealthy", "typehealthy - typecancer"))
## # A tibble: 60 × 14
## cell_group parameter factor c_lower c_effect c_upper c_pH0 c_FDR v_lower
## <chr> <chr> <chr> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl>
## 1 B immature typecanc… <NA> -1.89 -1.34 -0.780 0 0 NA
## 2 B immature typeheal… <NA> 0.780 1.34 1.89 0 0 NA
## 3 B mem typecanc… <NA> -2.25 -1.63 -0.988 0 0 NA
## 4 B mem typeheal… <NA> 0.988 1.63 2.25 0 0 NA
## 5 CD4 cm S10… typecanc… <NA> -1.46 -0.986 -0.527 0 0 NA
## 6 CD4 cm S10… typeheal… <NA> 0.527 0.986 1.46 0 0 NA
## 7 CD4 cm hig… typecanc… <NA> 0.808 1.55 2.25 0 0 NA
## 8 CD4 cm hig… typeheal… <NA> -2.25 -1.55 -0.808 0 0 NA
## 9 CD4 cm rib… typecanc… <NA> 0.375 0.980 1.60 0.00250 5.31e-4 NA
## 10 CD4 cm rib… typeheal… <NA> -1.60 -0.980 -0.375 0.00250 5.31e-4 NA
## # ℹ 50 more rows
## # ℹ 5 more variables: v_effect <dbl>, v_upper <dbl>, v_pH0 <dbl>, v_FDR <dbl>,
## # count_data <list>
This is achieved through model comparison with loo
. In the following
example, the model with association with factors better fits the data
compared to the baseline model with no factor association. For
comparisons check_outliers
must be set to FALSE as the leave-one-out
must work with the same amount of data, while outlier elimination does
not guarantee it.
If elpd_diff
is away from zero of > 5 se_diff
difference of 5, we
are confident that a model is better than the other
reference.
In this case, -79.9 / 11.5 = -6.9, therefore we can conclude that model
one, the one with factor association, is better than model two.
library(loo)
# Fit first model
model_with_factor_association =
seurat_obj |>
sccomp_estimate(
formula_composition = ~ type,
.sample = sample,
.cell_group = cell_group,
bimodal_mean_variability_association = TRUE,
inference_method = "hmc",
enable_loo = TRUE
)
# Fit second model
model_without_association =
seurat_obj |>
sccomp_estimate(
formula_composition = ~ 1,
.sample = sample,
.cell_group = cell_group,
bimodal_mean_variability_association = TRUE,
inference_method = "hmc",
enable_loo = TRUE
)
# Compare models
loo_compare(
attr(model_with_factor_association, "fit")$loo(),
attr(model_without_association, "fit")$loo()
)
We can model the cell-group variability also dependent on the type, and so test differences in variability
res =
seurat_obj |>
sccomp_estimate(
formula_composition = ~ type,
formula_variability = ~ type,
.sample = sample,
.cell_group = cell_group,
bimodal_mean_variability_association = TRUE,
cores = 1, verbose = FALSE
)
res
## # A tibble: 60 × 10
## cell_group parameter factor c_lower c_effect c_upper v_lower v_effect v_upper
## <chr> <chr> <chr> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl>
## 1 B immature (Interce… <NA> 0.350 0.759 1.17 -4.37 -3.94 -3.48
## 2 B immature typeheal… type 0.768 1.34 1.87 -1.03 -0.289 0.330
## 3 B mem (Interce… <NA> -1.35 -0.832 -0.351 -5.13 -4.58 -4.07
## 4 B mem typeheal… type 0.992 1.67 2.37 -1.69 -0.789 -0.0238
## 5 CD4 cm S1… (Interce… <NA> 1.35 1.68 2.01 -3.80 -3.36 -2.91
## 6 CD4 cm S1… typeheal… type 0.364 0.811 1.25 -1.34 -0.752 -0.243
## 7 CD4 cm hi… (Interce… <NA> -1.05 -0.511 0.0150 -5.20 -4.67 -4.17
## 8 CD4 cm hi… typeheal… type -1.94 -0.936 0.0524 0.531 1.55 2.57
## 9 CD4 cm ri… (Interce… <NA> -0.158 0.301 0.773 -5.10 -4.57 -4.04
## 10 CD4 cm ri… typeheal… type -1.73 -1.03 -0.362 -0.0762 0.530 1.30
## # ℹ 50 more rows
## # ℹ 1 more variable: count_data <list>
We recommend setting bimodal_mean_variability_association = TRUE
. The
bimodality of the mean-variability association can be confirmed from the
plots$credible_intervals_2D (see below).
We recommend setting bimodal_mean_variability_association = FALSE
(Default).
It is possible to directly evaluate the posterior distribution. In this example, we plot the Monte Carlo chain for the slope parameter of the first cell type. We can see that it has converged and is negative with probability 1.
library(cmdstanr)
## This is cmdstanr version 0.8.1.9000
## - CmdStanR documentation and vignettes: mc-stan.org/cmdstanr
## - CmdStan path: /Users/a1234450/.cmdstan/cmdstan-2.35.0
## - CmdStan version: 2.35.0
library(posterior)
## This is posterior version 1.6.0
##
## Attaching package: 'posterior'
## The following objects are masked from 'package:stats':
##
## mad, sd, var
## The following objects are masked from 'package:base':
##
## %in%, match
library(bayesplot)
## This is bayesplot version 1.11.1
## - Online documentation and vignettes at mc-stan.org/bayesplot
## - bayesplot theme set to bayesplot::theme_default()
## * Does _not_ affect other ggplot2 plots
## * See ?bayesplot_theme_set for details on theme setting
##
## Attaching package: 'bayesplot'
## The following object is masked from 'package:posterior':
##
## rhat
# Assuming res contains the fit object from cmdstanr
fit <- res %>% attr("fit")
# Extract draws for 'beta[2,1]'
draws <- as_draws_array(fit$draws("beta[2,1]"))
# Create a traceplot for 'beta[2,1]'
mcmc_trace(draws, pars = "beta[2,1]")
Plot 1D significance plot
plots = res |> sccomp_test() |> plot()
## Loading model from cache...
## Running standalone generated quantities after 1 MCMC chain, with 1 thread(s) per chain...
##
## Chain 1 finished in 0.0 seconds.
## Joining with `by = join_by(cell_group, sample)`
## Joining with `by = join_by(cell_group, type)`
plots$credible_intervals_1D
Plot 2D significance plot. Data points are cell groups. Error bars are the 95% credible interval. The dashed lines represent the default threshold fold change for which the probabilities (c_pH0, v_pH0) are calculated. pH0 of 0 represent the rejection of the null hypothesis that no effect is observed.
This plot is provided only if differential variability has been tested.
The differential variability estimates are reliable only if the linear
association between mean and variability for (intercept)
(left-hand
side facet) is satisfied. A scatterplot (besides the Intercept) is
provided for each category of interest. For each category of
interest, the composition and variability effects should be generally
uncorrelated.
plots$credible_intervals_2D
The new tidy framework was introduced in 2024, two, understand the differences and improvements. Compared to the old framework, please read this blog post.