elasticNetSMLcv {sparseSEM} | R Documentation |
While elasticNetSML function has a set of default (alpha, lambda) and the optimal one is chosen by 5 fold cv, elasticNetSMLcv tests the combination of a set of alpha an lambda, and choose one as the optimal parameters. elasticNetSMLcv should be combined with elasticNetSMLpoint to obtain the network inference. For each alpha from the set of alphas provided, the function perform 5 fold CV for each user supplied lambda to determine the optimal alpha and lambda for the data.
elasticNetSMLcv(Y, X, Missing, B, alpha_factors,lambda_factors, Verbose)
Y |
gene expression M by N matrix |
X |
cis_eQTL M by N matrix |
Missing |
missing data in Y |
B |
true network topology if available |
alpha_factors |
alpha_factors: the set of alphas to be tested, and is in range of (0, 1); |
lambda_factors |
penalty lambda_factor: the set of lambda to be tested, and is in range of (0, 1); |
Verbose |
describe the information output from 0 - 10, larger number means more output |
the function perform CV and parameter inference, calculate power and FDR
Bout |
the matrix B from sparseSEM |
fout |
f: the weight for matrix X |
stat |
compute the power and FDR statistics if the ture topology is provided |
simTime |
computational time |
residual |
only meaningful for 1 alpha: |
Difference in three functions:
1) elasticNetSML: Default alpha = 0.95: -0.05: 0.05; default 20 lambdas
2) elasticNetSMLcv: user supplied alphas (one or more), lambdas; compute the optimal parameters and network parameters
3) elasticNetSMLpoint: user supplied one alpha and one lambda, compute the network parameters
Anhui Huang; Dept of Electrical and Computer Engineering, Univ of Miami, Coral Gables, FL
1. Cai, X., Bazerque, J.A., and Giannakis, G.B. (2013). Inference of Gene Regulatory Networks with Sparse Structural Equation Models Exploiting Genetic Perturbations. PLoS Comput Biol 9, e1003068.
2. Huang A., Bazerque J. A., Giannakis G. B., DUroso G., Myers C. L., Cai X., Elastic Net algorithm for inferring gene regulatory networks based on structural equation models, to be submitted.
library(sparseSEM) data(B); data(Y); data(X); data(Missing); OUT <- elasticNetSMLcv(Y, X, Missing, B, alpha_factors = c(0.75, 0.5, 0.25), lambda_factors=c(0.1, 0.01, 0.001), Verbose = 1);