multi_omic_clustering: Multi-omic clustering via multi-view clustering or...

In vittoriofortino84/COPS: Clustering algorithms for Omics-driven Patient Stratification

Multi-omic clustering via multi-view clustering or integration

Description

Multi-omic clustering via multi-view clustering or integration

Usage

multi_omic_clustering(
  dat_list,
  meta_data = NULL,
  multi_omic_methods = "ANF",
  n_clusters = 2,
  distance_metric = "euclidean",
  correlation_method = "spearman",
  standardize_data = FALSE,
  non_negativity_transform = rep_len("none", length(dat_list)),
  view_distributions = rep_len("gaussian", length(dat_list)),
  icp_lambda = rep(0.03, length(dat_list)),
  icp_burnin = 100,
  icp_draw = 200,
  icp_maxiter = 20,
  icp_sdev = 0.05,
  icp_eps = 1e-04,
  icb_burnin = 1000,
  icb_draw = 1200,
  icb_sdev = 0.5,
  icb_thin = 1,
  nmf_maxiter = 200,
  nmf_st.count = 20,
  nmf_n.ini = 30,
  nmf_ini.nndsvd = TRUE,
  nmf_scaling = "F-ratio",
  mofa_convergence_mode = "medium",
  mofa_maxiter = 1000,
  mofa_environment = NULL,
  mofa_lib_path = NULL,
  anf_neighbors = 20,
  kkmeans_algorithm = "spectral",
  kkmeans_refine = FALSE,
  kkmeans_maxiter = 100,
  kkmeans_n_init = 100,
  kkmeans_tol = 1e-08,
  mkkm_mr_lambda = 1,
  mkkm_mr_tolerance = 1e-08,
  mkkm_mr_mosek = FALSE,
  mkkm_mr_mosek_verbosity = 1L,
  ecmc_a = 1,
  ecmc_b = 1,
  ecmc_eps = 1e-06,
  ecmc_maxiter = 100,
  ecmc_mkkm_mr = TRUE,
  data_is_kernels = FALSE,
  zero_var_removal = TRUE,
  mvc_threads = 1,
  gene_id_list = NULL,
  preprocess_data = TRUE,
  ...
)

Arguments

dat_list	List of input data.frames for input.
meta_data	A single data.frame or a list that includes meta data for each view. If a list is provided, at the moment only the first element is used (by appending to clustering output).
multi_omic_methods	Vector of algorithm names to be applied. See details.
n_clusters	Integer vector of number of clusters to output.
distance_metric	Distance metric for clustering factorized data (only for MOFA).
correlation_method	Correlation method for distance_metric, if applicable.
standardize_data	If set, standardizes data before clustering.
non_negativity_transform	Vector of transformation names for IntNMF. See details below.
view_distributions	A vector specifying the distribution to use for each view. Used by iCluster+, iClusterBayes and MOFA2. Options are "gaussian", "bernoulli" and "poisson".
icp_lambda	iCluster+ L1 penalty for each view. See iClusterPlus.
icp_burnin	iCluster+ number of MCMC burn in samples for approximating joint distribution of latent variables. See iClusterPlus.
icp_draw	iCluster+ number of MCMC samples to draw after burn in for approximating joint distribution of latent variables. See iClusterPlus.
icp_maxiter	iCluster+ maximum number of Newton-Rhapson (EM) iterations. See iClusterPlus. iClusterBayes.
icp_sdev	iCluster+ MCMC random walk standard deviation. See iClusterPlus.
icp_eps	iCluster+ algorithm convergence threshold. See iClusterPlus.
icb_burnin	iClusteBayes number of samples for MCMC burn in. See iClusterBayes.
icb_draw	iClusteBayes number of MCMC samples to draw after burn in. See iClusterBayes.
icb_sdev	iClusteBayes MCMC random walk standard deviation. See iClusterBayes.
icb_thin	iClusteBayes MCMC thinning, only one sample in every icb_thin samples will be used. See iClusterBayes.
nmf_maxiter	Maxiter for IntNMF. See nmf.mnnals.
nmf_st.count	Count stability for IntNMF. See nmf.mnnals.
nmf_n.ini	Number of initializations for IntNMF. See nmf.mnnals.
nmf_ini.nndsvd	If set, IntNMF uses NNDSVD for initialization. See nmf.mnnals.
nmf_scaling	Omic weights that are used for scaling. Defaults to the Frobenius norm ratio similarly to Chalise et al. 2017.
mofa_convergence_mode	MOFA convergence threshold. See get_default_training_options.
mofa_maxiter	MOFA maximum iterations. See get_default_training_options.
mofa_environment	If set, uses the specified Python environment (with mofapy). Defaults to basilisk.
mofa_lib_path	Path to libpython. May be required if using non-default mofa_environment.
anf_neighbors	Number of neighbours to use in knn-graph.
kkmeans_algorithm	See kernel_kmeans.
kkmeans_refine	See kernel_kmeans.
kkmeans_maxiter	See kernel_kmeans.
kkmeans_n_init	See kernel_kmeans.
kkmeans_tol	See kernel_kmeans.
mkkm_mr_lambda	Regularization parameter for mkkm_mr.
mkkm_mr_tolerance	Convergence threshold for mkkm_mr.
mkkm_mr_mosek	If set, uses Rmosek for convex optimization instead of CVXR for mkkm_mr.
mkkm_mr_mosek_verbosity	MOSEK verbosity parameter for mkkm_mr.
ecmc_a	Regularization parameter for ECMC.
ecmc_b	Regularization parameter for ECMC.
ecmc_eps	Convergence threshold for ECMC.
ecmc_maxiter	Maximum number of iterations for ECMC.
ecmc_mkkm_mr	If set, uses mkkm_mr on consensus kernels obtained from ECMC. Otherwise uses the average kernel and kernel k-means.
data_is_kernels	If TRUE, input data is assumed to be kernel matrices. Otherwise kernels are computed based on input data and the kernels parameter of get_multi_omic_kernels.
zero_var_removal	If set, removes all zero variance features from the data. It is called fold-wise, because this is assumed to be run inside CV.
mvc_threads	Number of threads to use for supported operations.
gene_id_list	List of gene/feature names for each view. If set, matches pipeline standardized feature names ("dim1", "dim2", ...) to names on the list. Required for pathway kernels.
preprocess_data	If the input data has already been processed by the COPS-pipeline, this should be disabled.
...	Arguments are passed to clustering_analysis when using MOFA and get_multi_omic_kernels when using kernel methods.

Details

Supported methods:

• "ANF" - Affinity Network Fusion ANF

• "iClusterPlus" or "iCluster+" - iClusterPlus. Supports only up to 4 views.

• "iClusterBayes" - codeiClusterBayes. Supports only up to 6 views

• "IntNMF" - Integrative Non-negative Matrix Factorization nmf.mnnals.

• "average_kernel" - kernel k-means with average kernel.

• "mkkm_mr" - Multiple Kernel K-Means with Matrix-induced Regularization mkkm_mr.

• "ECMC" - Enhanced Consensus Multi-view Clustering ECMC.

• "MOFA2" - Multi-Omics Factor Analysis. See vignette("getting_started_R", "MOFA2"). Resulting factorization is clustered with single-view algorithms by using clustering_analysis.

For supported kernels see get_multi_omic_kernels:

NMF non-negativity transform may be necessary if non-negativity was not considered while pre-processing the data. There are a few convenience functions included to transform the data as needed:

• "logistic" - 1/(1 + exp(-x)), maps input from (-Inf,Inf) to [0,1]. Used for e.g. microarray data or methylation M-values.

• "rank" - ranks values and divides by length, maps input from (-Inf,Inf) to [0,1].

• "offset2" - adds 2 to input. Useful for e.g. copy number alterations (assuming no alterations lower than -2).

Value

data.frame of clustering results

vittoriofortino84/COPS documentation built on Jan. 28, 2025, 3:16 p.m.