SAVER: Single-cell Analysis Via Expression Recovery

Description

The SAVER package implements SAVER, a gene expression recovery method for single-cell RNA sequencing (scRNA-seq). Borrowing information across all genes and cells, SAVER provides estimates for true expression levels as well as posterior distributions to account for estimation uncertainty. See Huang et al (2018) for more details.

Author(s)

Mo Huang (Maintainer), mohuangx@gmail.com

Nancy Zhang, nzh@wharton.upenn.edu

Mingyao Li, mingyao@pennmedicine.upenn.edu

Source

https://github.com/mohuangx/SAVER

Optimizes variance

Description

Finds the prior parameter that maximizes the marginal likelihood given the prediction.

Usage

calc.a(y, mu, sf)

calc.b(y, mu, sf)

calc.k(y, mu, sf)

Arguments

Details

calc.a returns a prior alpha parameter assuming constant coefficient of variation. calc.b returns a prior beta parameter assuming constant Fano factor. calc.k returns a prior variance parameter assuming constant variance.

Value

A vector with the optimized parameter and the negative log-likelihood.

Calculate estimate

Description

Calculates SAVER estimate

Usage

calc.estimate(
  x,
  x.est,
  cutoff = 0,
  coefs = NULL,
  sf,
  scale.sf,
  pred.gene.names,
  pred.cells,
  null.model,
  nworkers,
  calc.maxcor,
  estimates.only
)

calc.estimate.mean(x, sf, scale.sf, mu, nworkers, estimates.only)

calc.estimate.null(x, sf, scale.sf, nworkers, estimates.only)

Arguments

Details

The SAVER method starts by estimating the prior mean and variance for the true expression level for each gene and cell. The prior mean is obtained through predictions from a LASSO Poisson regression for each gene implemented using the glmnet package. Then, the variance is estimated through maximum likelihood assuming constant variance, Fano factor, or coefficient of variation variance structure for each gene. The posterior distribution is calculated and the posterior mean is reported as the SAVER estimate.

Value

A list with the following components

Calculates marginal likelihood

Description

Calculates the marginal likelihood given the prediction under constant coefficient of variation (a), Fano factor (b), and variance (k).

Usage

calc.loglik.a(a, y, mu, sf)

calc.loglik.b(b, y, mu, sf)

calc.loglik.k(k, y, mu, sf)

Arguments

Details

calc.loglik.a returns the shifted negative log-likelihood under constant coefficient of variation. calc.loglik.b returns the shifted negative log-likelihood under constant Fano factor. calc.loglik.k returns the shifted negative log-likelihood under constant variance.

Value

A shifted negative marginal log-likelihood.

Calculate maximum correlation

Description

Calculates the maximum absolute correlation between two matrices along the columns

Usage

calc.maxcor(x1, x2)

Arguments

Details

This function calculates the maximum absolute correlation for each column of x2 against each column of x1. The matrices are named and if the names overlap between x1 and x2, the correlation between the same named entries is set to zero.

Value

A vector of maximum absolute correlations

Examples

x1 <- matrix(rnorm(500), 100, 5)
x2 <- x1 + matrix(rnorm(500), 100, 5)
colnames(x1) <- c("A", "B", "C", "D", "E")
colnames(x2) <- c("A", "B", "C", "D", "E")
cor(x1, x2)
calc.maxcor(x1, x2)

Calculates SAVER posterior

Description

Given prediction and prior variance, calculates the Gamma posterior distribution parameters for a single gene.

Usage

calc.post(y, mu, sf, scale.sf)

Arguments

Details

Let \alpha be the shape parameter and \beta be the rate parameter of the prior Gamma distribution. Then, the posterior Gamma distribution will be

Gamma(y + \alpha, sf + \beta),

where y is the observed gene count and sf is the size factor.

Value

A list with the following components

Combines SAVER

Description

Combines SAVER objects

Usage

combine.saver(saver.list)

Arguments

Details

If SAVER was applied to a dataset for chunks of genes (by specifying pred.genes and pred.genes.only = TRUE), this function combines the individual SAVER objects into one SAVER object.

Value

A combined SAVER object

Examples

data("linnarsson")

## Not run: 
a <- saver(linnarsson, pred.genes = 1:5, pred.genes.only = TRUE)
b <- saver(linnarsson, pred.genes = 6:10, pred.genes.only = TRUE)
ab <- combine.saver(list(a, b))

## End(Not run)

Calculates gene-to-gene and cell-to-cell SAVER correlation

Description

Adjusts for SAVER estimation uncertainty by calculating and adjusting gene-to-gene and cell-to-cell correlation matrices

Usage

cor.genes(x, cor.mat = NULL)

cor.cells(x, cor.mat = NULL)

Arguments

Details

The SAVER estimates that are produced have varying levels of uncertainty depending on the gene and the cell. These functions adjust the gene-to-gene and cell-to-cell correlations of the SAVER estimates to reflect the estimation uncertainty.

Value

An adjusted correlation matrix.

Examples

data("linnarsson_saver")

gene.cor <- cor.genes(linnarsson_saver)

Calculates SAVER prediction.

Description

Uses cv.glmnet from the glmnet package to return the SAVER prediction.

Usage

expr.predict(
  x,
  y,
  pred.cells = 1:length(y),
  seed = NULL,
  lambda.max = NULL,
  lambda.min = NULL
)

Arguments

Details

The SAVER method starts with predicting the prior mean for each cell for a specific gene. The prediction is performed using the observed normalized gene count as the response and the normalized gene counts of other genes as predictors. cv.glmnet from the glmnet package is used to fit the LASSO Poisson regression. The model with the lowest cross-validation error is chosen and the fitted response values are returned and used as the SAVER prediction.

Value

A vector of predicted gene expression.

Output prior predictions

Description

Outputs prior predictions

Usage

get.mu(x, saver.obj)

Arguments

Details

This function outputs prior mean predictions \mu used in fitting the SAVER model.

Value

A matrix of prior mean predictions

Examples

data("linnarsson")
data("linnarsson_saver")

mu <- get.mu(linnarsson, linnarsson_saver)

Mouse brain single-cell RNA-seq dataset

Description

3,529 genes and 200 cells from a mouse brain scRNA-seq dataset.

Usage

linnarsson

Format

An object of class matrix with 3529 rows and 200 columns.

References

Zeisel, A., Munoz-Manchado, A. B., Codeluppi, S., Lonnerberg, P., La Manno, G., Jureus, A., ... Linnarsson, S. (2015). Cell types in the mouse cortex and hippocampus revealed by single-cell RNA-seq. Science, 347(6226), 1138-1142.

SAVER recovered mouse brain single-cell RNA-seq dataset

Description

Output of running 'saver' on the 'linnarsson' dataset.

Usage

linnarsson_saver

Format

An object of class saver of length 3.

References

Zeisel, A., Munoz-Manchado, A. B., Codeluppi, S., Lonnerberg, P., La Manno, G., Jureus, A., ... Linnarsson, S. (2015). Cell types in the mouse cortex and hippocampus revealed by single-cell RNA-seq. Science, 347(6226), 1138-1142.

Samples from SAVER

Description

Samples from the posterior distribution output by SAVER.

Usage

sample.saver(x, rep = 1, efficiency.known = FALSE, seed = NULL)

Arguments

Details

The SAVER method outputs a posterior distribution, which we can sample from for downstream analysis. The posterior distribution accounts for uncertainty in the SAVER estimation procedure. If the efficiency is known, negative binomial sampling is performed; otherwise, gamma sampling is performed.

Value

A matrix of expression values sampled from SAVER posterior. If rep > 1, a list of matrices is returned

Examples

data("linnarsson_saver")

samp1 <- sample.saver(linnarsson_saver, seed = 50)

Runs SAVER

Description

Recovers expression using the SAVER method.

Usage

saver(
  x,
  do.fast = TRUE,
  ncores = 1,
  size.factor = NULL,
  npred = NULL,
  pred.cells = NULL,
  pred.genes = NULL,
  pred.genes.only = FALSE,
  null.model = FALSE,
  mu = NULL,
  estimates.only = FALSE
)

Arguments

Details

The SAVER method starts by estimating the prior mean and variance for the true expression level for each gene and cell. The prior mean is obtained through predictions from a LASSO Poisson regression for each gene implemented using the glmnet package. Then, the variance is estimated through maximum likelihood assuming constant variance, Fano factor, or coefficient of variation variance structure for each gene. The posterior distribution is calculated and the posterior mean is reported as the SAVER estimate.

Value

If 'estimates.only = TRUE', then a matrix of SAVER estimates.

If 'estimates.only = FALSE', a list with the following components

The info element is a list with the following components:

Examples

data("linnarsson")

## Not run: 
system.time(linnarsson_saver <- saver(linnarsson, ncores = 12))

## End(Not run)

# predictions for top 5 highly expressed genes
## Not run: 
saver2 <- saver(linnarsson, npred = 5)

## End(Not run)

# predictions for certain genes
## Not run: 
genes <- c("Thy1", "Mbp", "Stim2", "Psmc6", "Rps19")
genes.ind <- which(rownames(linnarsson) 
saver3 <- saver(linnarsson, pred.genes = genes.ind)

## End(Not run)

# return only certain genes
## Not run: 
saver4 <- saver(linnarsson, pred.genes = genes.ind, pred.genes.only = TRUE)

## End(Not run)

Fits SAVER

Description

Fits SAVER object

Usage

saver.fit(
  x,
  x.est,
  do.fast,
  ncores,
  sf,
  scale.sf,
  pred.genes,
  pred.cells,
  null.model,
  ngenes = nrow(x),
  ncells = ncol(x),
  gene.names = rownames(x),
  cell.names = colnames(x),
  estimates.only
)

saver.fit.mean(
  x,
  ncores,
  sf,
  scale.sf,
  mu,
  ngenes = nrow(x),
  ncells = ncol(x),
  gene.names = rownames(x),
  cell.names = colnames(x),
  estimates.only
)

saver.fit.null(
  x,
  ncores,
  sf,
  scale.sf,
  ngenes = nrow(x),
  ncells = ncol(x),
  gene.names = rownames(x),
  cell.names = colnames(x),
  estimates.only
)

Arguments

Details

The SAVER method starts by estimating the prior mean and variance for the true expression level for each gene and cell. The prior mean is obtained through predictions from a Lasso Poisson regression for each gene implemented using the glmnet package. Then, the variance is estimated through maximum likelihood assuming constant variance, Fano factor, or coefficient of variation variance structure for each gene. The posterior distribution is calculated and the posterior mean is reported as the SAVER estimate.

Value

A list with the following components