Wrapper function for GPU Scrublet doublet detection parameters

GPU counterpart to bixverse::params_scrublet(). Two differences from the CPU list. The pca sub-list is gone: the GPU SVD is always randomised, so random_svd and sparse have nothing to switch and no_pcs is a plain argument. And the kNN block is backend dependent, see knn_backend.

Usage

params_scrublet_gpu(
  sim_doublet_ratio = 1.5,
  expected_doublet_rate = 0.1,
  stdev_doublet_rate = 0.02,
  n_bins_histogram = 100L,
  manual_threshold = NULL,
  no_pcs = 30L,
  normalisation = list(),
  hvg = list(),
  knn_backend = c("gpu", "cpu"),
  knn = list(k = 0L)
)

Arguments

sim_doublet_ratio

Numeric. Number of doublets to simulate relative to the number of observed cells. Defaults to 1.5.

expected_doublet_rate

Numeric in [0, 1]. Expected doublet rate, typically 0.05-0.10 depending on cell loading. Defaults to 0.1.

stdev_doublet_rate

Numeric in [0, 1]. Uncertainty in the expected doublet rate. Defaults to 0.02.

n_bins_histogram

Integer. Histogram bins for the Otsu threshold search. Defaults to 100L.

manual_threshold

Optional numeric. Fixed doublet score threshold. If NULL, Otsu's method picks it.

no_pcs

Integer. Number of principal components. Defaults to 30L.

normalisation

List. Optional overrides. See bixverse::params_norm_doublets_defaults() for the available parameters: log_transform, mean_center, normalise_variance, target_size.

hvg

List. Optional overrides. See bixverse::params_hvg_defaults() for the available parameters: min_gene_var_pctl, hvg_method, loess_span, clip_max, n_bins, binning_strategy.

knn_backend

String. One of "gpu" or "cpu". Picks which nearest neighbour index runs over the combined observed-plus-simulated embedding, and with it which keys knn accepts. "gpu" is the fast default; "cpu" buys the exact CPU indices at the cost of a host round trip on a matrix that is (1 + sim_doublet_ratio) * n_cells rows tall.

knn

List. Optional overrides for the kNN block. Validated against params_knn_gpu_defaults() when knn_backend = "gpu" and against bixverse::params_knn_defaults() when knn_backend = "cpu". Unknown keys are an error, not a silent pass-through. Defaults to list(k = 0L), which asks Rust to pick k.

Value

A flat named list with all GPU Scrublet parameters.

Details

Leave knn_method alone unless you have a reason. "exhaustive" is the default and is the right answer for Scrublet on the GPU arm.

Scrublet queries at a high k by construction: the count is taken over an embedding (1 + sim_doublet_ratio) * n_cells rows tall, and k = 0L then scales k by the same factor, so a 20k-cell run searches at k around 175. Exhaustive barely notices k, since the scan is the cost and k only sizes the top-k selection. NN-descent notices a lot: its build degree tracks k, so the descent does more work per node as k climbs. Measured at 20k cells and 30 PCs, exhaustive took 0.98s against 38.8s for NN-descent at k = 200. NN-descent only came out ahead at k = 10.

"ivf" is the one worth trying: it was the quickest of the three across that sweep and holds a Pearson above 0.99 against exhaustive.

Recall matters more here than elsewhere. The doublet score is a neighbour count, so a backend that drops neighbours biases every score downwards.

References

Wolock, et al., Cell Syst, 2020