umap

umap

umap(
  pc_scores,
  n_neighbors = 15,
  n_components = 2,
  method = c("uwot", "umap-learn"),
  metric = "euclidean",
  min_dist = 0.1,
  cores = 4,
  seed = -1,
  densemap = TRUE,
  dense_lambda = 2,
  verbose = FALSE,
  ...
)

Arguments

pc_scores

Matrix: exported from get_pc_scores()

n_neighbors

Integer: number of nearest neighbors to use

n_components

Integer: number of UMAP dimensions to calculate

method

Character: The backend method for calculating UMAP (either 'uwot' or 'umap-learn'). Default is 'uwot'.

metric

Character: How to calculate similarity. See details below for more informaation.

min_dist

Numeric: minimum distance to consider

cores

Integer: Number of cores to use for UMAP calculation

seed

Integer: reproducibility seed. If cores > 1 or seed equals -1, seed will be ignored.

densemap

Boolean: whether to use DensMAP for local densities

dense_lambda

Numeric: value to apply for local density. Default: 2. higher values prioritize local density while low values are closer to typical UMAP

verbose

Boolean: whether to be verbose in function calls

...

other parameters to pass to uwot::umap directly

Value

matrix with UMAP reductions

Details

method:

Currently, there are two ways that UMAP can be calculated: 'uwot' or 'umap-learn'. To use 'uwot', it's simply the installation of a dependency (likely already installed if FastPCA is installed). This is the same method that is default for Seurat, and is very efficient and fast. Some of the parameters that are explicit in the function call (like n_neighbors and n_components) are passed while others shown in uwot::umap's documentation can be passed by the ....

Alternatively, there is a python method called umap-learn. To use this, a conda environment is preferred. FastPCA provides interfaces to create and then activate the environment with FastPCA::setup_py_env() and FastPCA::start_FastPCA_env(). Something to keep in mind, that if using this method it's recommended (required?) to restart your R session, load FastPCA, then FastPCA::start_FastPCA_env(). There are system level conflicts somwhere between reticulate and R's torch package.

metric:

There are many metrics that are supported in the python implementation. Here are the list in the documentation for the umap function;

• Minkowski syle metrics: "euclidean", "manhattan", "chebyshev", "minkowski";

• Miscellaneous spatial metrics: "canberra", "braycurtis", "haversine";

• Normalized spatial metrics: "mahalanobis", "wminkowski", "seuclidean";

• Angular and correlation metrics: "cosine", "correlation";

• Metrics for binary data: "hamming", "jaccard", "dice", "russellrao", "kulsinski", "rogerstanimoto", "sokalmichener", "sokalsneath", "yule"