subtype_anf: Subtypes using Affinity Network Fusion
In agapow/subtypr: Robust and Validated Patient Subtyping for Precision Medicine

Description Usage Arguments Value See Also

Subtypes using Affinity Network Fusion

subtype_anf(data_list, minimal_return = FALSE, cluster_number, k_affi,
  alpha_affi = 1/6, beta_affi = 1/6, k_fusion = 20,
  weigth_fusion = NULL, type_fusion = c("two-step", "one-step"),
  alpha_fusion = c(1, 1, 0, 0, 0, 0, 0, 0), spectral_type = c("rw",
  "sym", "unnormalized"), verbose_fusion = FALSE, just_fuse = FALSE)

`data_list`	a list of data matrices with continuous data of format samples x features (with the same number of samples).
`minimal_return`	logical, if TRUE, the result of the function will just be what's needed to evaluate the goodness of the partition, i.e. the partition and the element for internal metrics.
`cluster_number`	The supposed or previously infered number of clusters.
`k_affi`	The number of k-nearest neighbors for the affinity matrix calculation.
`alpha_affi`	Coefficient for local diameters for affinity matrix calculation. Default value: 1/6.
`beta_affi`	Coefficient for pair-wise distance for affinity matrix calculation. Default value: 1/6.
`k_fusion`	the number of k nearest neighbors for function kNN_graph
`weigth_fusion`	a list of non-negative real numbers (which will be normalized internally so that it sums to 1) that one-to-one correspond to the affinity matrices computed from the features matrices provided in `data_list`. If not set, internally uniform weights are assigned to all the affinity matrices.
`type_fusion`	choose one of the two options "one-step" random walk, or "two-step" random walk (for the fusion process).
`alpha_fusion`	a list of eight non-negative real numbers (which will be normalized internally to make it sums to 1). Only used when "two-step" (default value of `type_fusion`) random walk is used. `alpha_fusion` is the weights for eight terms in the "two-step" random walk formula (check research paper for more explanations about the terms). Default value: (1, 1, 0, 0, 0, 0, 0, 0), i.e., only use the first two terms (since they are most effective in practice).
`spectral_type`	choose one of three versions of graph Laplacian: "unnormalized": unnormalized graph Laplacian matrix (L = D - W`); "rw": normalization closely related to random walk (L = I - D^(-1)W)`; (default choice) "sym": normalized symmetric matrix (L = I - D^(-0.5) W * D^(-0.5)') For more information: https://www.cs.cmu.edu/~aarti/Class/10701/readings/Luxburg06_TR.pdf
`verbose_fusion`	logical(1); if true, print some information concerning the fusion step.
`just_fuse`	wetherto just integrate matrices and return the fused matrix or not

a result list containing:

$partition: The predicted partition
$element_for_metric: The name of the element in the result list. containing the data to be used with internal metrics.
$affinity_fused: The fused affinity matrix returned by the function ANF

ANF, affinity_matrix, spectral_clustering.

agapow/subtypr documentation built on May 5, 2019, 1:33 a.m.