snpnet: Fit the Lasso/Elastic-Net for Large Phenotype-Genotype...
In junyangq/GWASLasso: snpnet: Efficient Lasso Solver for Large SNP Data
snpnet R Documentation
Fit the Lasso/Elastic-Net for Large Phenotype-Genotype Datasets
Description
Fit the entire lasso or elastic-net solution path using the Batch Screening Iterative Lasso (BASIL) algorithm
on large phenotype-genotype datasets.
Usage
snpnet(genotype.pfile, phenotype.file, phenotype, family = NULL, covariates = NULL, alpha
= 1, nlambda = 100, lambda.min.ratio = ifelse(nobs < nvars, 0.01, 1e-04), lambda = NULL,
split.col = NULL, p.factor = NULL, status.col = NULL, mem = NULL, configs = NULL)
Arguments
genotype.pfile
the PLINK 2.0 pgen file that contains genotype.
We assume the existence of genotype.pfile.pgen,pvar.zst,psam.
phenotype.file
the path of the file that contains the phenotype values and can be read as
as a table. There should be FID (family ID) and IID (individual ID) columns
containing the identifier for each individual, and the phenotype column(s).
(optional) some covariate columns and a column specifying the
training/validation split can be included in this file.
phenotype
the name of the phenotype. Must be the same as the corresponding column name in
the phenotype file.
family
the type of the phenotype: "gaussian", "binomial", or "cox". If not provided or NULL,
it will be detected based on the number of levels in the response.
covariates
a character vector containing the names of the covariates included in the lasso
fitting, whose coefficients will not be penalized. The names must exist in the
column names of the phenotype file.
alpha
the elastic-net mixing parameter, where the penalty is defined as
alpha * ||beta||_1 + (1-alpha)/2 * ||beta||_2^2. alpha = 1 corresponds to the lasso penalty,
while alpha = 0 corresponds to the ridge penalty.
nlambda
the number of lambda values - default is 100.
lambda.min.ratio
smallest value for lambda, as a fraction of lambda.max, the (data derived) entry value,
i.e. the smallest value for which all coefficients are zero. The default
depends on the sample size nobs relative to the number of actual variables
nvars (after QC filtering). If nobs > nvars, the default is 0.0001, close to zero.
If nobs < nvars, the default is 0.01. A very small value of lambda.min.ratio
will lead to a saturated fit in the nobs < nvars case.
lambda
one can specify the full lambda list on which the lasso/elastic-net will be solved.
Once provided, 'lambda' and 'lambda.min.ratio' will be ignored. It can be used for refitting
after the optimal parameter is selected by validation.
split.col
the column name in the phenotype file that specifies the membership of individuals to
the training or the validation set. The individuals marked as "train" and "val" will
be treated as the training and validation set, respectively. When specified, the
model performance is evaluated on both the training and the validation sets.
p.factor
a named vector of separate penalty factors applied to each coefficient. This is
a number that multiplies lambda to allow different shrinkage. If not provided, default is 1
for all variables. Otherwise should be complete and positive for all variables.
status.col
the column name for the status column for Cox proportional hazards model.
When running the Cox model, the specified column must exist in the phenotype file.
mem
Memory (MB) 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.
configs
a list of other config parameters.
- missing.rate
variants are excluded if the missing rate exceeds this level. Default is 0.1.
- MAF.thresh
variants are excluded if the minor allele frequency (MAF) is lower
than this level. Default is 0.001.
- nCores
the number of cores used for computation. You may use the maximum number
of cores available on the computer. Default is 1, single core.
- num.snps.batch
the number of variants added to the strong set in each iteration. Default is 1000.
- niter
The number of maximum iteration in the algorithm. Note that each iteration
may be able to find solutions for more than one lambda value. The default is 50
- prevIter
if non-zero, it indicates the last successful iteration in the procedure so that
we can restart from there. niter should be no less than prevIter.
- save
a logical value whether to save the intermediate results (e.g. in case of job failure and restart).
- results.dir
the path to the directory where meta and intermediate results are saved.
- meta.dir
the relative path to the subdirectory used to store the computed
summary statistics, e.g. mean, missing rate, standard deviation (when 'standardization = TRUE').
Needed when 'save = TRUE'. Default is '"meta.dir/'.
- save.dir
the relative path to the subdirectory used to store the intermediate
results so that we may look into or recover from later.
Needed when 'save = TRUE'. Default is '"results/'.
- excludeSNP
character vector containing genotype names to exclude from
the analysis
- nlams.init
the number of lambdas considered in the first iteration.
Default 10 is a reasonable number to start with.
- nlams.delta
the length of extended lambdas down the sequence when there are few
left in the current sequence (remember we don't fit all lambdas
every iteration, only extend when most of the current ones have been completed and validated). Default is 5.
- glmnet.thresh
the convergence threshold used in glmnet/glmnetPlus.
- keep
one may specify keep file in plink format to focus on a subset of individuals.
- use.glmnetPlus
a logical value whether to use glmnet with warm start, if
the glmnetPlus package is available. Currently only "gaussian" family is supported.
- early.stopping
a logical value indicating whether early stopping based on validation metric is desired.
- stopping.lag
a parameter for the stopping criterion such that the procedure stops after
this number of consecutive decreases in the validation metric.
- verbose
a logical value indicating if more detailed messages should be printed.
- KKT.verbose
a logical value indicating if details on KKT check should be printed.
- increase.size
the increase in batch size if the KKT condition fails often in recent iterations.
Default is half of the batch size.
- plink2.path
the user-specified path to plink2 (default: plink2)
- zstdcat.path
the user-specified path to zstdcat (default: zstdcat)
- zcat.path
the user-specified path to zcat (to read a zcat compressed phenotype file) (default: zdcat)
- rank
if TRUE, then the smallest lambda indices when each variable enters the model are recorded
Details
Junyang Qian, Wenfei Du, Yosuke Tanigawa, Matthew Aguirre, Robert Tibshirani, Manuel A. Rivas, and Trevor Hastie.
"A Fast and Flexible Algorithm for Solving the Lasso in Large-scale and Ultrahigh-dimensional Problems."
bioRxiv (2019): https://doi.org/10.1101/630079
Value
A list containing the solution path, the metric evaluated on training/validation set and others.
junyangq/GWASLasso documentation built on Nov. 25, 2022, 10:03 a.m.
R Package Documentation
Browse R Packages
We want your feedback!
Note that we can't provide technical support on individual packages. You should contact the package authors for that.
| snpnet | R Documentation |
Fit the Lasso/Elastic-Net for Large Phenotype-Genotype Datasets
Description
Fit the entire lasso or elastic-net solution path using the Batch Screening Iterative Lasso (BASIL) algorithm on large phenotype-genotype datasets.
Usage
snpnet(genotype.pfile, phenotype.file, phenotype, family = NULL, covariates = NULL, alpha = 1, nlambda = 100, lambda.min.ratio = ifelse(nobs < nvars, 0.01, 1e-04), lambda = NULL, split.col = NULL, p.factor = NULL, status.col = NULL, mem = NULL, configs = NULL)
Arguments
genotype.pfile |
the PLINK 2.0 pgen file that contains genotype. We assume the existence of genotype.pfile.pgen,pvar.zst,psam. |
phenotype.file |
the path of the file that contains the phenotype values and can be read as as a table. There should be FID (family ID) and IID (individual ID) columns containing the identifier for each individual, and the phenotype column(s). (optional) some covariate columns and a column specifying the training/validation split can be included in this file. |
phenotype |
the name of the phenotype. Must be the same as the corresponding column name in the phenotype file. |
family |
the type of the phenotype: "gaussian", "binomial", or "cox". If not provided or NULL, it will be detected based on the number of levels in the response. |
covariates |
a character vector containing the names of the covariates included in the lasso fitting, whose coefficients will not be penalized. The names must exist in the column names of the phenotype file. |
alpha |
the elastic-net mixing parameter, where the penalty is defined as alpha * ||beta||_1 + (1-alpha)/2 * ||beta||_2^2. alpha = 1 corresponds to the lasso penalty, while alpha = 0 corresponds to the ridge penalty. |
nlambda |
the number of lambda values - default is 100. |
lambda.min.ratio |
smallest value for lambda, as a fraction of lambda.max, the (data derived) entry value, i.e. the smallest value for which all coefficients are zero. The default depends on the sample size nobs relative to the number of actual variables nvars (after QC filtering). If nobs > nvars, the default is 0.0001, close to zero. If nobs < nvars, the default is 0.01. A very small value of lambda.min.ratio will lead to a saturated fit in the nobs < nvars case. |
lambda |
one can specify the full lambda list on which the lasso/elastic-net will be solved. Once provided, 'lambda' and 'lambda.min.ratio' will be ignored. It can be used for refitting after the optimal parameter is selected by validation. |
split.col |
the column name in the phenotype file that specifies the membership of individuals to the training or the validation set. The individuals marked as "train" and "val" will be treated as the training and validation set, respectively. When specified, the model performance is evaluated on both the training and the validation sets. |
p.factor |
a named vector of separate penalty factors applied to each coefficient. This is a number that multiplies lambda to allow different shrinkage. If not provided, default is 1 for all variables. Otherwise should be complete and positive for all variables. |
status.col |
the column name for the status column for Cox proportional hazards model. When running the Cox model, the specified column must exist in the phenotype file. |
mem |
Memory (MB) 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. |
configs |
a list of other config parameters.
|
Details
Junyang Qian, Wenfei Du, Yosuke Tanigawa, Matthew Aguirre, Robert Tibshirani, Manuel A. Rivas, and Trevor Hastie. "A Fast and Flexible Algorithm for Solving the Lasso in Large-scale and Ultrahigh-dimensional Problems." bioRxiv (2019): https://doi.org/10.1101/630079
Value
A list containing the solution path, the metric evaluated on training/validation set and others.
R Package Documentation
Browse R Packages
We want your feedback!
Note that we can't provide technical support on individual packages. You should contact the package authors for that.
Embedding an R snippet on your website
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.