In DeLorean: Estimates Pseudotimes for Single Cell Expression Data

# Attach the DeLorean data frame to access members
attach(dl)

Posterior

Pseudotime

The estimated pseudotime from the best sample against the observed capture time.

plot(dl, type="pseudotime")

The full tau posterior for each cell.

(
    ggplot(samples.l$tau %>% arrange(capture),
           aes(x=cell, y=tau, color=capture))
    + geom_boxplot()
    + coord_flip()
)

The posterior of the pseudotime offsets relative to the capture times. We also show the prior on the offset.

plot(dl, type="tau.offsets")

Between time variation

The posterior for the between time variation $\log \psi_g$:

(ggplot(samples.l$psi, aes(x=log(psi)))
    + geom_density()
    + geom_rug()
    + stat_function(fun=function(x) dnorm(x,
                                          mean=hyper$mu_psi,
                                          sd=hyper$sigma_psi),
                    colour="blue", alpha=.7, linetype="dashed")
)

Temporal variation by gene:

(ggplot(noise.levels,
        aes(x=gene, y=log10(psi), fill=gene %in% genes.high.psi),
        environment=environment())
    + geom_boxplot()
    + coord_flip())

Within time variation

The posterior for the within time variation $\log \omega_g$:

(ggplot(samples.l$omega, aes(x=log(omega)))
    + geom_density()
    + geom_rug()
    + stat_function(fun=function(x) dnorm(x,
                                          mean=hyper$mu_omega,
                                          sd=hyper$sigma_omega),
                    colour="blue", alpha=.7, linetype="dashed")
)

Within time variation by gene:

(ggplot(noise.levels,
        aes(x=gene, y=log10(omega), fill=gene %in% genes.high.psi),
        environment=environment())
    + geom_boxplot()
    + coord_flip())

Variation comparisons

The within time variation compared to the temporal variation (on log-log scale):

(ggplot(noise.levels,
        aes(x=log(psi), y=log(omega), color=gene %in% genes.high.psi),
        environment=environment())
    + geom_point()
    + geom_abline(intercept=0, slope=1, linetype="dashed", alpha=.7))

have.cell.sizes <- 'S' %in% names(samples.l)

Other posteriors

The posterior for the model log probability:

(ggplot(samples.l$lp__, aes(x=lp__))
    + geom_density()
    + geom_rug())

# Detach the previously attached DeLorean data frame
detach(dl)

DeLorean documentation built on May 2, 2019, 9:24 a.m.