tune_lucid {LUCIDus}

A wrapper function to perform model selection for LUCID

Description

Given a grid of K and L1 penalties (incluing Rho_G, Rho_Z_mu and Rho_Z_Cov; for LUCID early only), fit LUCID model over all combinations of K and L1 penalties to determine the optimal penalty. Note that the input of the grid of K differs for different LUCID models. i.e. For LUCID Early, K = 3:5; for LUCID in parallel, K = list(2:3, 2:3); for LUCID in serial, K = list(list(2:3,2),2:3)

Usage

tune_lucid(
  G,
  Z,
  Y,
  CoG = NULL,
  CoY = NULL,
  family = c("normal", "binary"),
  K,
  lucid_model = c("early", "parallel", "serial"),
  Rho_G = 0,
  Rho_Z_Mu = 0,
  Rho_Z_Cov = 0,
  verbose_tune = FALSE,
  ...
)

Arguments

G

Exposures, a numeric vector, matrix, or data frame. Categorical variable should be transformed into dummy variables. If a matrix or data frame, rows represent observations and columns correspond to variables.

Z

Omics data, if "early", an N by M matrix; If "parallel", a list, each element i is a matrix with N rows and P_i features; If "serial", a list, each element i is a matrix with N rows and p_i features or a list with two or more matrices with N rows and a certain number of features

Y

Outcome, a numeric vector. Categorical variable is not allowed. Binary outcome should be coded as 0 and 1.

CoG

Optional, covariates to be adjusted for estimating the latent cluster. A numeric vector, matrix or data frame. Categorical variable should be transformed into dummy variables.

CoY

Optional, covariates to be adjusted for estimating the association between latent cluster and the outcome. A numeric vector, matrix or data frame. Categorical variable should be transformed into dummy variables.

family

Distribution of outcome. For continuous outcome, use "normal"; for binary outcome, use "binary". Default is "normal".

K

Number of latent clusters. If "early", an integer;If "parallel",a list, each element is an integer/integer vector, same length as Z; If "serial", a list, each element is either an integer or an list of integers, same length as Z. If K is given as a grid, the input of the grid of K differs for different LUCID models. i.e. For LUCID Early, K = 3:5; for LUCID in parallel, K = list(2:3, 2:3); for LUCID in serial, K = list(list(2:3,2),2:3)

lucid_model

Specifying LUCID model, "early" for early integration, "parallel" for lucid in parallel, "serial" for lucid in serial

Rho_G

A scalar or a vector. This parameter is the LASSO penalty to regularize exposures. If it is a vector, tune_lucid will conduct model selection and variable selection. User can try penalties from 0 to 1. Work for LUCID early only.

Rho_Z_Mu

A scalar or a vector. This parameter is the LASSO penalty to regularize cluster-specific means for omics data (Z). If it is a vector, tune_lucid will conduct model selection and variable selection. User can try penalties from 1 to 100. Work for LUCID early only.

Rho_Z_Cov

A scalar or a vector. This parameter is the graphical LASSO penalty to estimate sparse cluster-specific variance-covariance matrices for omics data (Z). If it is a vector, tune_lucid will conduct model selection and variable selection. User can try penalties from 0 to 1. Work for LUCID early only.

verbose_tune

A flag to print details of tuning process.

...

Other parameters passed to estimate_lucid

Value

A list:

best_model

the best model over different combination of tuning parameters

tune_list

a data frame contains combination of tuning parameters and c orresponding BIC

res_model

a list of LUCID models corresponding to each combination of tuning parameters

Examples

## Not run: 
# use simulated data
G <- sim_data$G
Z <- sim_data$Z
Y_normal <- sim_data$Y_normal

# find the optimal model over the grid of K
tune_K <- tune_lucid(G = G, Z = Z, Y = Y_normal, lucid_model = "early",
 useY = FALSE, tol = 1e-2,
seed = 1, K = 2:3)

# tune penalties
tune_Rho_G <- tune_lucid(G = G, Z = Z, Y = Y_normal, lucid_model = "early",
 useY = FALSE, tol = 1e-2,
seed = 1, K = 2, Rho_G = c(0.1, 0.2))
tune_Rho_Z_Mu <- tune_lucid(G = G, Z = Z, Y = Y_normal, lucid_model = "early", 
useY = FALSE, tol = 1e-2,
seed = 1, K = 2, Rho_Z_Mu = c(10, 20))
tune_Rho_Z_Cov <- tune_lucid(G = G, Z = Z, Y = Y_normal, lucid_model = "early", 
useY = FALSE, tol = 1e-2,
seed = 1, K = 2, Rho_Z_Cov = c(0.1, 0.2))

## End(Not run)