Association tests yield weird results in single lineage case

[SONGDONGYUAN1994]

Hi,
Thanks for this nice package! I am trying to apply tradeSeq to single lineage simulation data, but the results seem to be incorrect.
Here is my code:

suppressMessages({library(devtools)
library(tidyverse, quietly = TRUE)
library(TSCAN)
library(plyr)
library(slingshot, quietly = TRUE)
library(RColorBrewer)
library(SingleCellExperiment)
library(scales)
library(gridExtra)
library(splatter)
library(dyntoy)
library(dynwrap)
library(mgcv)
library(mclust, quietly = TRUE)
library(tradeSeq, quietly = TRUE)
library(dynplot)
library(destiny, quietly = TRUE)})

set.seed(12)

dataset <- generate_dataset(
  model = model_linear(num_milestones = 2),
  num_cells = 500,
  num_features = 5000,
 differentially_expressed_rate = 0.01,
  normalise = T
)

counts <- dataset$counts %>% as.matrix() %>% t()
colnames(counts) <- dataset$cell_ids
norms <- dataset$expression %>% t()

simu_dat <- SingleCellExperiment(assays = List(counts = counts, norms = norms))

pca <- prcomp(t(assays(simu_dat)$norms), scale. = FALSE)
rd <- pca$x[,1:3]

reducedDims(simu_dat) <- SimpleList(PCA = rd)
cl <- Mclust(rd, G = 4)$classification
colData(simu_dat)$GMM <- cl

simu_slingshot <- slingshot(simu_dat, clusterLabels = 'GMM', reducedDim = 'PCA')

lin <- getLineages(simu_dat, clusterLabels = colData(simu_slingshot)$GMM)

crv <- getCurves(lin) %>% SlingshotDataSet()

norms <- assays(simu_slingshot)$norms %>% as.matrix()

sce <- fitGAM(counts = counts, pseudotime = slingPseudotime(crv, na = FALSE),
                  sds = crv, nknots = 3, sce = FALSE)

assoRes <- associationTest(sce)
assoRes %>% as_tibble(rownames = "gene") %>% dplyr::arrange(desc(waldStat))

The results show that all genes are significant, and the Wald statistics are extremely high, which does not make sence. Would you mind check this? Thanks!

[koenvandenberge]

Thank you for letting us know.
It seems like mgcv is not respecting the number of knots that are provided as input. Even though we're telling it to fit a smoother with 3 knots, it fits a smoother with only 2 knots, which sends the associationTest astray since it thinks that the model has 3 knots. Interestingly, this only ever happens when only a single lineage is fitted.
We will look into the best way to solve this.

[koenvandenberge]

Hi, this should now be fixed.
Thanks again for letting us know about this issue, and please let us know if you encounter any further issues.

Best,
Koen

[SONGDONGYUAN1994]

Thanks guys!

[cstrlln]

I'm actually having this issue while running a single lineage, any suggestion with the number of knots?