v.pudms: test a PU fit on a test data set

In RomeroLab/pudms: Positive-Unlabeled Learning for the Analysis of Deep Mutational Scanning Datasets

Description

test a PU fit on a test data set

Usage

v.pudms(
  protein_dat,
  py1 = NULL,
  nhyperparam = 10,
  nfolds = 5,
  test_idx = 1:nfolds,
  seed = round(runif(1, min = 1, max = 1000)),
  order = 1,
  refstate = NULL,
  verbose = T,
  nobs_thresh = 10,
  lambda = 0,
  pvalue = FALSE,
  n_eff_prop = 1,
  intercept = F,
  maxit = 1000,
  eps = 0.001,
  inner_eps = 0.01,
  initial_coef = NULL,
  p.adjust.method = "BH",
  tol = 1e-05,
  nCores = 1,
  full.fit = FALSE,
  full.fit.pvalue = FALSE,
  outfile = NULL
)

Arguments

protein_dat	input data. A data table containing (sequence, labeled, unlabeled, seqId)
py1	a numeric value, a numeric vector or NULL; the prevalence of positives in the unlabeled data. If length(py1) >1, optimal py1 will be chosen based on auc values on a test data set. If NULL (default), a sequence of py1 values (of length nhyperparam)–ranging from 0.001 to 0.5 interpolated in a log scale–will be considered.
nhyperparam	an integer for the length of the py1 sequence if py1 == NULL
nfolds	the number of subsamples. (nfolds -1)/nfolds splits will be used for training, and the rest will be used for testing.
test_idx	a vector of indices of cross-validation models which will be fitted. Default is to fit the model for each of the cross-validation fold.
seed	a seed number for reproducibility
order	an integer; 1= main effects, 2= main effects + pairwise effects
refstate	a character which will be used for the common reference state; the default is to use the most frequent amino acid as the reference state for each of the position.
verbose	a logical value. The default is TRUE
nobs_thresh	the number of minimum required mutations per position
lambda	l1 penalty
pvalue	a logial value; if TRUE, p-values based on the asymptotic distribution are obtained
n_eff_prop	proportion of an effective sample size
intercept	a logical value; if TRUE, an estimated intercept is reported together with other coefficients
maxit	maximum number of iterations
eps	convergence threshold for the outer loop
inner_eps	convergence threshold for the inner loop
initial_coef	a vector representing an initial point where we start PUlasso algorithm from.
p.adjust.method	method for multiple comparison
tol	NULL or a numeric value; if the estimated roc curve <= y+tol, the estimated roc curve is determined to be contained by the maximal curve. The default is NULL, where we use tol = 1sd value of the length(test_idx) roc curves at each x value of the estimated roc curve.
nCores	the number of threads for computing.
full.fit	a logical value; if TRUE, the model will be fitted using a full data set and at a chosen py1.
full.fit.pvalue	a logical value; if TRUE, p-values for the full fit will be returned
outfile	NULL or a string; if a string is provided, an output with the name of the string will be exported in a working directory.

Value

a list containing v.dmsfit (all fits using training/test splits), roc_curves (average roc curve at each py1), dmsfit (pudms.fit using a full data set at the selected py1), folds (test/training split information), py1 (a sequence of py1 values used for searching), py1.opt (the selected py1 value based on the predictive performance of the models)

RomeroLab/pudms documentation built on Jan. 2, 2021, 5:10 a.m.