Finds the shared nearest neighbourhood for the cells. This supplies a graph.

Neighbourhood graph generator utilised. According to the assay name a different method will be used. If the assay is derived from scanpy then the scanpy algorithm will be applied. Otherwise, the seurat implementation will be applied.

perform.nn(
  object,
  assay,
  reduction,
  neighbour.name.suffix = "",
  dims.use = NULL,
  k.param = 20,
  prune.SNN = 1/15,
  nn.method = "annoy",
  n.trees = 50,
  nn.eps = 0,
  annoy.metric = "euclidean",
  n_neighbors = 15,
  random_state = 0,
  method = "umap",
  metric = "euclidean",
  generate.diffmap = FALSE,
  n_comps = 15,
  diffmap.name.suffix = "",
  verbose = FALSE,
  seed = 1234
)

Arguments

object

IBRAP S4 class object

assay

Character. String containing indicating which assay to use

reduction

Character. String defining which reduction to supply to the clustering algorithm.

neighbour.name.suffix

Character. String defining the names to store the neighbourhood graph results under.

dims.use

Numerical. (scanpy only) How many components of the reduction should be used, 0 means that all will be used. Default = 0

k.param

Numerical. The number of k when calcuating k-nearest neighbour. Default = 20

prune.SNN

Numerical. Setas acceptance cutoff for jaccard index whilst computing neighbourhood overlap for SNN construction. Any edges with a value less than this parameter will be removed. 0 = no pruning and 1 = prune everything. Default = 0

nn.method

Character. Nearest neighbour method, either 'rann' or 'annoy'. Default = 'annoy'

n.trees

Numerical. More trees facilitates hgiher precision when using 'annoy' method. Default = 20

nn.eps

Numerical. Margin of error when performing nearest neighbour search whilst using rann method. 0 would imply an exact search. Default = 0.0

annoy.metric

Character. Distance metric for annoy method. Options: 'euclidean', 'cosine', 'manhattan', 'hamming'. Default = 'euclidean'

n_neighbors

Numerical. (scanpy only) How many neighbours should be found per cell, a higher value typically achieves more accurate results. Default = 15

random_state

Numerical. (scanpy only) The seed value to use. Default = 0

method

Character. (scanpy only) String indicating which methodology to use including: ‘umap’, ‘gauss’ or ‘rapids’

metric

Character. (scanpy only) String indicating which distance metric to use, including: ‘braycurtis’, ‘canberra’, ‘chebyshev’, ‘correlation’, ‘dice’, ‘hamming’, ‘jaccard’, ‘kulsinski’, ‘mahalanobis’, ‘minkowski’, ‘rogerstanimoto’, ‘russellrao’, ‘seuclidean’, ‘sokalmichener’, ‘sokalsneath’, ‘sqeuclidean’, ‘yule’, ‘cityblock’, ‘cosine’, ‘euclidean’, ‘l1’, ‘l2’ or ‘manhattan’

generate.diffmap

Boolean. (scanpy only) Should diffusion maps be generated from the neighourhood graphs, these will be stored in computational_reductions and can be used for umap generation and further neighbourhood generation. Default = TRUE

n_comps

Numerical. (scanpy only) How many components should be generated for the diffusion maps. Default = 15

compute.SNN

Boolean. Should the shared nearest neighbour graph be calculated. Default = TRUE

diffmap.name

Character. (scanpy only) What should the diffusion maps be named.

Examples

# generates a diffusion map from the scanpy assay
object <- perform.nn(object = object, assay = c('SCT', 'SCRAN', 'SCANPY'), 
                     reduction = c('pca'),
                     dims = list(0,0))
#> Error in is(object, "IBRAP"): object 'object' not found