multisnpnet: Fast Multi-Phenotype SRRR on SNP Data

In junyangq/multiSnpnet: Fit a sparse reduced rank regression model on large-scale SNP data and multivariate responses

Fast Multi-Phenotype SRRR on SNP Data

Description

Fit a sparse reduced rank regression model on large-scale SNP data and multivariate responses with batch variable screening and alternating minimization. It computes a full solution path on a grid of penalty values. Can deal with larger-than-memory SNP data, missing values and adjustment covariates.

Usage

multisnpnet(genotype_file, phenotype_file, phenotype_names, binary_phenotypes = NULL,
  covariate_names, rank, nlambda = 100, lambda.min.ratio = 0.01, standardize_response = TRUE,
  weight = NULL, validation = FALSE, split_col = NULL, mem = NULL,
  batch_size = 100, prev_iter = 0, max.iter = 10, configs = NULL, save = TRUE,
  early_stopping = FALSE)

Arguments

genotype_file	Path to the suite of genotype files. genotype_file.pgen, psam, pvar.zst must exist.
phenotype_file	Path to the phenotype. The header must include FID, IID, covariate_names and phenotype_names. Missing values are expected to be encoded as -9.
binary_phenotypes	Names of the binary phenotypes. AUC will be evaluated for binary phenotypes.
covariate_names	Character vector of the names of the adjustment covariates.
rank	Target rank of the model.
nlambda	Number of penalty values.
lambda.min.ratio	Ratio of the minimum penalty to the maximum penalty.
standardize_response	Boolean. Whether to standardize the responses before fitting to deal with potential different units of the responses.
weight	Numberic vector that specifies the (importance) weights for the responses.
p.factor	Named vector of separate penalty factors applied to each coefficient. This is a number that multiplies lambda to allow different shrinkage. Default is 1 for all variables. Can specify partially and the rest will be set to 1. Must be positive.
validation	Boolean. Whether to evaluate on validation set.
split_col	Name of the column in the phenotype file that specifies whether each sample belongs to the training split or the validation split. The values are either "train" or "val".
mem	Memory available for the program. It tells PLINK 2.0 the amount of memory it can harness for the computation. IMPORTANT if using a job scheduler.
batch_size	Number of variants used in batch screening.
prev_iter	Index of the iteration to start from (e.g. to resume a previously interrupted computation).
max.iter	Maximum number of iterations allowed for alternating minimization.
configs	List of additional configuration parameters. It can include: nCores number of cores for the PLINK computation (default: 1) results.dir directory to save intermediate results if save=TRUE (default: temp directory created by the tempdir function) thresh convergence threshold for alternating minimization (default: 1E-7) glmnet.thresh convergence threshold for glmnet(Plus) (default: 1E-7) plink2.path path to the PLINK2.0 program, if not on the system path zstdcat.path path to the zstdcat program, if not on the system path use_safe whether to use safe product to deal with very large matrix multiplication (default: TRUE). One may also specify MAXLEN (default: (2^31-1)/2), the maximum vector length passed to the R base matrix multiplication operation excludeSNP character vector containing genotype names to exclude from the analysis
save	Boolean. Whether to save intermediate results.
early_stopping	Whether to stop the process early if validation metric starts to fall.
early_stopping_phenotypes	List of phenotypes to focus when evaluating the early stopping condition.
early_stopping_check_average	whether to check the average metric when evaluating the early stopping condition

junyangq/multiSnpnet documentation built on Oct. 19, 2023, 8:22 p.m.