Wrapper function for Palantir parameters

Parameters controlling the Palantir trajectory inference. The kNN graph you hand to run_palantir_sc() feeds the diffusion kernel. The geodesics are measured over a second kNN graph that Palantir builds internally on the multiscale space, and knn is what controls that one. It is a different knob from k in the kNN parameter block, which sizes the backend index. Palantir overrides k and ann_dist for its internal search, so only knn_method and the backend tuning parameters have an effect.

Usage

params_sc_palantir(
  n_dcs = 10L,
  n_eigs = NULL,
  knn = 30L,
  num_waypoints = 1200L,
  scale_components = TRUE,
  use_early_cell_as_start = TRUE,
  max_iterations = 25L,
  branch_prob_threshold = 0.01,
  lanczos_basis_size = NULL,
  lanczos_max_restarts = 16L,
  lanczos_tol = 1e-08,
  knn_params = list()
)

Arguments

n_dcs

Integer. Diffusion components to extract before the multiscale scaling. Defaults to 10L.

n_eigs

Integer or NULL. Eigenvectors to retain, not components: the trivial leading eigenvector is counted here and then dropped, so 3L leaves two multiscale components. If NULL, the count is picked from the largest eigengap, as the reference does. Defaults to NULL.

knn

Integer. Neighbours for the geodesic graph over the multiscale space, in the reference's self-inclusive convention. Defaults to 30L.

num_waypoints

Integer. Target waypoint count for the max-min sampler. Defaults to 1200L.

scale_components

Boolean. Min-max scale each multiscale component to [0, 1] before any distance is taken. Defaults to TRUE.

use_early_cell_as_start

Boolean. Use the provided early cell directly rather than snapping it to the nearest diffusion-map boundary cell. Defaults to TRUE.

max_iterations

Integer. Iteration cap for the pseudotime refinement. Defaults to 25L.

branch_prob_threshold

Numeric. Fate probabilities below this are zeroed. Defaults to 0.01.

lanczos_basis_size

Integer or NULL. Krylov basis vectors held at once during the diffusion-map eigendecomposition. If NULL, derived from the requested component count. Defaults to NULL.

lanczos_max_restarts

Integer. Maximum restart cycles for the Lanczos solver. Defaults to 16L.

lanczos_tol

Numeric. Relative residual tolerance for the Lanczos solver. Defaults to 1e-08.

knn_params

List. Optional overrides for the kNN parameters of the internal multiscale search. See params_knn_defaults() for available parameters: k, knn_method, ann_dist, search_budget, n_trees, delta, diversify_prob, ef_budget, m, ef_construction, ef_search, n_list and n_probe. See params_knn_defaults() for the available elements. Defaults to list().

Value

A named list with the following elements:

• n_dcs - Integer. Diffusion components to extract before the multiscale scaling. Defaults to 10L.

• n_eigs - Integer or NULL. Eigenvectors to retain, not components: the trivial leading eigenvector is counted here and then dropped, so 3L leaves two multiscale components. If NULL, the count is picked from the largest eigengap, as the reference does. Defaults to NULL.

• knn - Integer. Neighbours for the geodesic graph over the multiscale space, in the reference's self-inclusive convention. Defaults to 30L.

• num_waypoints - Integer. Target waypoint count for the max-min sampler. Defaults to 1200L.

• scale_components - Boolean. Min-max scale each multiscale component to [0, 1] before any distance is taken. Defaults to TRUE.

• use_early_cell_as_start - Boolean. Use the provided early cell directly rather than snapping it to the nearest diffusion-map boundary cell. Defaults to TRUE.

• max_iterations - Integer. Iteration cap for the pseudotime refinement. Defaults to 25L.

• branch_prob_threshold - Numeric. Fate probabilities below this are zeroed. Defaults to 0.01.

• lanczos_basis_size - Integer or NULL. Krylov basis vectors held at once during the diffusion-map eigendecomposition. If NULL, derived from the requested component count. Defaults to NULL.

• lanczos_max_restarts - Integer. Maximum restart cycles for the Lanczos solver. Defaults to 16L.

• lanczos_tol - Numeric. Relative residual tolerance for the Lanczos solver. Defaults to 1e-08.

• The elements of params_knn_defaults(), overridden by knn_params, spliced in at this position.

References

Setty, et al., Nat. Biotechnol., 2019.