Package {fspls2}

Type: Package
Title: A Package to Find Minimal Transcriptional Signatures
Version: 0.1.0
Date: 2026-09-02
Author: Lachlan Coin [aut, cre]
Imports: R6, jsonlite, Matrix, glmnet, tidyr, MASS, ggplot2, pROC, binom, confintr, DBI, RSQLite, methods,tibble, ggrepel, RColorBrewer, data.table
Depends: R (≥ 4.1.0)
Maintainer: Lachlan Coin <l.coin@imb.uq.edu.au>
Description: Identifies minimal biomarker signatures for predicting phenotypes from multiomic data. Integrates multiple omics layers, supports internal cross-validation. Supports missing values in predictors and outcomes. Enables model based imputation of uncertain values. Supports continuous, binary, multi-class and ordinal outcomes.
License: GPL-2
LazyLoad: yes
NeedsCompilation: no
Repository: CRAN
Packaged: 2026-09-02 04:43:34 UTC; lcoin
Date/Publication: 2026-09-12 12:30:02 UTC
Config/roxygen2/version: 8.0.0
Encoding: UTF-8
Suggests: knitr, rmarkdown, testthat
VignetteBuilder: knitr

Analysis Base Class

Description

A class that encapsulates a dataset holder and/or analysis of datasets

Details

Available methods:

Methods

Public methods

analysisBase$new()

Create a new instance of base class

Usage
analysisBase$new(nme, dims = NULL, flags = list(), dbDir = tempdir())
Arguments
nme

name

dims

dims

flags

a list object specifying options

dbDir

location for database to be written, default to tempdir

analysisBase$name()

get dataset name;

Usage
analysisBase$name()
Returns

assigned name of dataset

analysisBase$integrate()

integrates variables from different CV runs

Usage
analysisBase$integrate(variables)
Arguments
variables

a variables object

analysisBase$clone()

The objects of this class are cloneable with this method.

Usage
analysisBase$clone(deep = FALSE)
Arguments
deep

Whether to make a deep clone.

Analysis Class

Description

A class that encapsulates analysis of one or more datasets. Only expert users wlil need access to this

Details

Available methods:

@export

Super class

analysisBase -> analysisEnv

Methods

Public methods

Inherited methods

analysisEnv$new()

Create a new instance

Usage
analysisEnv$new(flags = list(), dbDir = tempdir())
Arguments
flags

a list object specifying options

dbDir

location for database to be written. If NULL then no DB gets the trajectory

analysisEnv$plot_trajectory()

Usage
analysisEnv$plot_trajectory(datasH)
Arguments
datasH

a list of dataH objects after variable selection selection for a single fold

analysisEnv$select_k()

Usage
analysisEnv$select_k(datasH, k1, vars_l_todo)
Arguments
datasH

a list of dataH objects

k1

the fold

vars_l_todo

an object representing what is left to do

Returns

vars_l_todo object Get todo

analysisEnv$getTodo()

Usage
analysisEnv$getTodo(flags, phens)
Arguments
flags

list of options

phens

phenotypes

Returns

object outlining what is left to do for next iteration Clear the databases. Only necessary if using a database and re-running analyses

analysisEnv$clear_db()

Usage
analysisEnv$clear_db(drop = FALSE, exclude = "vars", datasH = NULL)
Arguments
drop

completely drop tables and start from scratch

exclude

tables to exclue from clearing

datasH

list of dataH objects to clear Get the next vars in iteration. Internal function

analysisEnv$savePvalsAndNextVars()

Usage
analysisEnv$savePvalsAndNextVars(vars_l_todo, comb2_new, data_nme, k1)
Arguments
vars_l_todo

vars_l_todo

comb2_new

results

data_nme

name of dataset

k1

which repetition

Returns

object outlining what is left to do main function for variable selection

analysisEnv$select()

Usage
analysisEnv$select(
  datasH,
  phens,
  transform_x,
  data_types = datasH[[1]]$data_types()
)
Arguments
datasH

a list of dataH objects

phens

list of phenotypes

transform_x

transformation object

data_types

data_types object

analysisEnv$clone()

The objects of this class are cloneable with this method.

Usage
analysisEnv$clone(deep = FALSE)
Arguments
deep

Whether to make a deep clone.

Data holder class

Description

A class that encapsulates a dataset holder

Details

Available methods:

Super class

analysisBase -> dataH

Methods

Public methods

Inherited methods

dataH$new()

Create a new instance

Usage
dataH$new(
  data,
  y,
  certainty = rep(1, nrow(y)),
  nme,
  flags,
  family = getFamily(y),
  dbDir = NULL
)
Arguments
data

a list of matrices (of different modalities), with columns as variables and rows as samples

y

phenotype matrix, with columns as outcomes and rows as samples

certainty

the certainty of each observation, expressed as a probability of being correct. Default is to treat each observation as certain

nme

The name of the data objetct.

flags

list of options, described in the vignette

family

the statistical family of phenotype y, can be calculated by getFamily

dbDir

dir for database to store results to speed up re-reruns. Can be NULL

transform_x

a transformation object from the function getTransformation

dataH$multiAnglesAndPv()

Calculate the angles and pv across multiple phenotypes. This is an internal function and should not need to be called by user

Usage
dataH$multiAnglesAndPv(comb20, k1, expt_id, vars_l_todo)
Arguments
comb20

values from previous iteration

k1

k1 is the current fold

expt_id

and ID for the experiment (can be 0)

vars_l_todo

the variables under consideration clear database. This is only used if you want to clear the database. Internal function

phens1

phenotypes being used

flags

list of options

dataH$clear_db()

Usage
dataH$clear_db(drop = FALSE, exclude = "vars")
Arguments
drop

drop the tables instead of clear

exclude

which tables not to clear

dataH$get_trajectory()

returns data on trajectory

Usage
dataH$get_trajectory()
Returns

a dataframe with information to plot trajectory get the data types

dataH$data_types()

Usage
dataH$data_types(
  data_nmes = list(all = names(private$data$data)),
  inds = lapply(data_nmes, function(x) list(pos_inds = c(), neg_inds = c(), max = 1000)),
  direction = lapply(inds, private$data$getDirection),
  incl = c(),
  excl = c()
)
Arguments
data_nmes

a list of names of the data names to include, optional

inds

a list of positive and negative inds, optional

direction

a direction object

incl

a list of variables to include

excl

a list of variables to exclude

Returns

an object which is used to specify both directionallity and data types

dataH$split()

split dataset into smaller datasets

Usage
dataH$split(proportions = c(0.5, 0.5))
Arguments
proportions

what proportions to split into

Returns

a list of dataH objects with data partiioned according to proportions

dataH$pheno()

get the phenotyeps

Usage
dataH$pheno()
Returns

phenotypes

dataH$dims()

get the dimensions

Usage
dataH$dims()
Returns

dimensions get the nreps

dataH$nreps()

Usage
dataH$nreps()
Returns

nreps

dataH$update()

update the phenotypes without remaking the entire object. You can provide many phenotypes in the initialisation stage, but only consider a subset in model fitting stage in this way.Must run this before select

Usage
dataH$update(
  phens = self$pheno()$all,
  flags = private$flags,
  transform_x = fromJSON(flags$transform_x),
  data_types = self$data_types()
)
Arguments
phens

phenotypes

flags

list of options

transform_x

transformation object

data_types

a data_type object

dataH$select()

main function for variable selection

Usage
dataH$select(analysis = analysisEnv$new(flags = private$flags, dbDir = NULL))
Arguments
analysis

an analysisEnv object

dataH$y()

provides access to internal storage of phenotype data

Usage
dataH$y(phens = self$pheno()[[1]])
Arguments
phens

list of phenotypes

Returns

list of matrices

dataH$plotPredictions()

ggplot to visualise predictions

Usage
dataH$plotPredictions(all_modelsh, update = FALSE, liab = TRUE)
Arguments
all_modelsh

fitted models from makeAllModels

update

whether to update

liab

return liability score, or probability (for binomial, ordinal multinomial)

Returns

a ggplot

dataH$extractPredictions()

extract the predictions for the fitted models

Usage
dataH$extractPredictions(all_modelsh, liab = TRUE)
Arguments
all_modelsh

fitted models from makeAllModels

liab

return liability score, or probability (for binomial, ordinal multinomial)

Returns

a table with results

dataH$plotData()

plot all data for selected variables

Usage
dataH$plotData(
  variables,
  all_types = FALSE,
  violin = FALSE,
  assoc = FALSE,
  update = FALSE
)
Arguments
variables

object returned by fspls.select

all_types

use all types?

violin

violin plots

assoc

use association

update

whether to use attributes from bariables to update this object, default is FALSE

Returns

a table with results

dataH$updateWeights()

update the weights based on predictions from the model

Usage
dataH$updateWeights(all_models)
Arguments
all_models

the output of dataH function getAllModels

Returns

list of auc beta and updated sumdiff of weights used as stopping criteria

dataH$getYNew()

return updated y after iterative imputation

Usage
dataH$getYNew()
Returns

updated y, updated certainty and indices which were updated

dataH$getProjectedData()

get data after projection

Usage
dataH$getProjectedData(varnames)
Arguments
varnames

varnames

Returns

projected data after projecting out varnames

dataH$getVariance()

get variance of data

Usage
dataH$getVariance(varnames)
Arguments
varnames

varnames

Returns

variance

dataH$makeAllModels()

fit models based on variables

Usage
dataH$makeAllModels(variables, update = FALSE)
Arguments
variables

list of variables selected by select method

update

whether to automatically update phens, transform_x and flags , default TRUE

Returns

fitted models

dataH$evaluateAllModels()

evaluate the fit of models

Usage
dataH$evaluateAllModels(all_modelsh, update = TRUE)
Arguments
all_modelsh

models fitted from makeAllModels

update

whether to automatically update phens, transform_x and flags , default TRUE

Returns

evaluation of models

dataH$clone()

The objects of this class are cloneable with this method.

Usage
dataH$clone(deep = FALSE)
Arguments
deep

Whether to make a deep clone.

main function for variable selection with unknown values

Description

main function for variable selection with unknown values

Usage

fspls.iterative(dataset, flags, transform_x)

Arguments

dataset

a list with y and data value

flags

list of options

transform_x

transformation object

main function for variable selection

Description

main function for variable selection

Usage

fspls.select(
  datasH,
  flags,
  transform_x,
  phens = datasH[[1]]$pheno()$all,
  data_types = datasH[[1]]$data_types(),
  dbDir = NULL
)

Arguments

datasH

a list of dataH objects

flags

list of options

transform_x

transformation object

phens

list of phenotypes

data_types

a data type object

dbDir

database dir, can be NULL

Infer the statistical families of the phenotypes

Description

Infer the statistical families of the phenotypes

Usage

getFamily(y_mat, max_ordinal = getOption("max_ordinal", 10))

Arguments

y_mat

a matrix of phenotypes, with columns as variables and rows as samples

max_ordinal

maximum number of values before we consider an ordinal value as a continuous value

Value

a list with the statistical families detected

extracts the full model variables from a cross validation run

Description

extracts the full model variables from a cross validation run

Usage

getFolds(leng, nrep = NA, batch = NA, randomize = TRUE)

Arguments

leng

length of dataset

nrep

number of reps

batch

batch size (one of batch or nrep is non NA)

randomize

whether to randomize

Get object for transforming the x variables

Description

Get object for transforming the x variables

Usage

getTransform(
  pows = c(1),
  offset = 0,
  n_random = 1,
  perm = FALSE,
  norm = 1,
  CHECK = FALSE
)

Arguments

pows

power to raise to

offset

offset to subtract

n_random

how many random variables

perm

whether random is permutation

norm

rescaling factor

CHECK

check whether inverse function works

Value

transformation object

Plot the results from evaluateAll

Description

Plot the results from evaluateAll

Usage

plotEval(
  eval3,
  shape_color = c("pheno", "subpheno"),
  shape = shape_color,
  text = "variable",
  dotsize = "nsamps",
  color = shape_color,
  linetype = shape,
  showranges = TRUE,
  txtsize = 1,
  logy = FALSE,
  legend = FALSE,
  sep_by = "",
  scales = "free",
  point = TRUE,
  line = TRUE,
  labelsize = 2,
  grid0 = c("cohort", "measure"),
  grid1 = "cv_full",
  title = "",
  title1 = ""
)

Arguments

eval3

a tibble or data frame from evaluateAll method from dataH

shape_color

a list of column names to encode as both shape and color

shape

defaults to shape_color but can be specified separately

text

a list of column names to include as text labels

dotsize

a numerical column to encode dotsize, or a number for constant sizes

color

defaults to shape_color but can be specified separately

linetype

column controlling the type of line

showranges

whether to show 95% CI as ribbon

txtsize

size of text

logy

whether to log y axis

legend

whether to show a legend

sep_by

list of column names to separate into diff plots

scales

used in faceting can be free or fixed

point

whether to show dots

line

whether to include line

labelsize

the size of the text labels

grid0

how to facet on y

grid1

how to facet on x

title

the title

title1

columns to include in the title

Value

ggplot2 plot