Wrapper function for gene trend parameters
params_sc_gene_trends.RdParameters controlling the landmark Gaussian process that
run_gene_trends_sc() fits per branch. The kernel is a Matern-5/2
one and the prediction grid doubles as the landmark set. The defaults come
from the reference and are prior-dominated. Palantir's pseudotime is min-max
scaled to [0, 1], so a length_scale of 1.0 spans the entire domain and
a sigma of 1.0 sits at roughly the signal scale of log-normalised
expression. The posterior will flatten genuine transient structure and
resolve almost any gene into a smooth monotone or single-peaked curve. That
is a presentation choice, not inference. Shorten length_scale before
believing a bump.
Usage
params_sc_gene_trends(
resolution = 500L,
weighting = c("hard_mask", "fate_probability"),
length_scale = 1,
sigma = 1,
jitter = 1e-06,
max_jitter_retries = 3L,
chunk_size = 2048L
)Arguments
- resolution
Integer. Grid points per branch. Kept at the default even when a branch holds fewer cells, as the reference does. Defaults to
500L.- weighting
String. With
"hard_mask"every selected cell enters its branch's fit with equal weight, which is what the reference does. With"fate_probability"every cell enters every fit weighted by its fate probability, which is more defensible: a cell at 0.6 is not a member. One ofc("hard_mask", "fate_probability"). Defaults to"hard_mask".- length_scale
Numeric. Matern-5/2 length scale. Defaults to
1.0.- sigma
Numeric. Noise standard deviation. Defaults to
1.0.- jitter
Numeric. Added to the landmark covariance diagonal before the Cholesky. Defaults to
1e-06.- max_jitter_retries
Integer. Times the jitter is raised and the Cholesky retried before giving up. Defaults to
3L.- chunk_size
Integer. Training points held at once when accumulating the cross-covariance. Defaults to
2048L.
Value
A named list with the following elements:
resolution - Integer. Grid points per branch. Kept at the default even when a branch holds fewer cells, as the reference does. Defaults to
500L.weighting - String. With
"hard_mask"every selected cell enters its branch's fit with equal weight, which is what the reference does. With"fate_probability"every cell enters every fit weighted by its fate probability, which is more defensible: a cell at 0.6 is not a member. One ofc("hard_mask", "fate_probability"). Defaults to"hard_mask".length_scale - Numeric. Matern-5/2 length scale. Defaults to
1.0.sigma - Numeric. Noise standard deviation. Defaults to
1.0.jitter - Numeric. Added to the landmark covariance diagonal before the Cholesky. Defaults to
1e-06.max_jitter_retries - Integer. Times the jitter is raised and the Cholesky retried before giving up. Defaults to
3L.chunk_size - Integer. Training points held at once when accumulating the cross-covariance. Defaults to
2048L.