The dsGamlssClient package is a
DataSHIELD client-side package that
includes the server-side functions to fit Generalized Additive Models
for Location, Scale and Shape (GAMLSS) [1] using DataSHIELD. It is
based on the original
gamlss implementation [1]
and the dsBaseClient
package [2].
DataSHIELD is a software infrastructure which allows you to do
non-disclosive federated analysis on sensitive data. The DataSHIELD
website has in depth descriptions of what
it is, how it works and how to install it. A key point to highlight is
that DataSHIELD has a client-server infrastructure, so the
dsGamlssClient package needs to be used in conjunction with the
dsGamlss package - trying to use
one without the other makes no sense. Detailed instructions on how to
install DataSHIELD can be found at the DataSHIELD
Wiki. Discussion and help with using
DataSHIELD can be obtained from the DataSHIELD
Forum.
You can install the dsGamlssClient package from
GitHub with:
# install.packages("devtools")
devtools::install_github("bips-hb/dsGamlssClient")To successfully run the package, the dsBase and dsGamlss server-side packages must be installed on each DataSHIELD server. Instructions on how to install a server-side package on a DataSHIELD server can be found in the Data Manager Section at the DataSHIELD Wiki.
The example uses the server-less DataSHIELD implementation
DSLite [3] to illustrate
the use of the package’s main functions ds.gamlss and
ds.predict.gamlss. Thus, access to a DataSHIELD server is not required
to follow the example.
First, you need to install the DSLite package, which is available on
CRAN.
install.packages("DSLite", repos = "https://cloud.r-project.org/")Note that the DSLite package is only required if you want to use the
server-less DataSHIELD implementation. If instead you are using
Armadillo or
Opal DataSHIELD servers you must
to install the
DSMolgenisArmadillo
or DSOpal package from CRAN
with install.packages("DSMolgenisArmadillo") or
install.packages("DSOpal").
Furthermore, to follow the example, the server-side packages dsBase
and dsGamlss must be installed from GitHub in the global environment.
Again, this is only necessary for the server-less DataSHIELD
implementation. If you are using Armadillo or Opal DataSHIELD servers,
you must install the dsBase and dsGamlss packages on the server, as
described e.g. in the Data Manager
Section
at the DataSHIELD Wiki.
# install.packages("devtools")
devtools::install_github("datashield/dsBase")
devtools::install_github("bips-hb/dsGamlss")
library(dsGamlssClient)
library(DSLite)
library(dsBase)
library(dsGamlss)
data(mtcars)For illustrative purposes, the mtcars example data is split across two
servers. Therefore, two DSLite servers are set up, with the required
server-side packages dsBase and dsGamlss. Furthermore, each server
holds a subset of the mtcars data. Analyzing the data on the two
servers jointly with ds.gamlss, is mathematically equivalent to
fitting a GAMLSS model to the whole mtcars data. Note however, that if
nonparametric terms are included in the model, there might be slight
numerical differences between ds.gamlss and gamlss::gamlss, since
the matrix equation to obtain the regression coefficients is solved
differently.
dslite.server1 <- newDSLiteServer(
tables = list(data = mtcars[c(1:15), ]),
config = defaultDSConfiguration(include = c("dsBase", "dsGamlss"))
)
dslite.server2 <- newDSLiteServer(
tables = list(data = mtcars[c(16:nrow(mtcars)), ]),
config = defaultDSConfiguration(include = c("dsBase", "dsGamlss"))
)
builder <- DSI::newDSLoginBuilder()
builder$append(server = "study1", url = "dslite.server1", table = "data", driver = "DSLiteDriver")
builder$append(server = "study2", url = "dslite.server2", table = "data", driver = "DSLiteDriver")
logindata.dslite <- builder$build()
# Login to the virtualized server
conns <- DSI::datashield.login(logindata.dslite, assign = TRUE)
#>
#> Logging into the collaborating servers
#>
#> No variables have been specified.
#> All the variables in the table
#> (the whole dataset) will be assigned to R!
#>
#> Assigning table data...
DSI::datashield.assign.table(conns = conns, symbol = "D", table = c("data", "data"))Then, one can apply the ds.gamlss function to fit a GAMLSS model. In
this case, it is assumed that the minimum and maximum of wt, which are
used to determine the knots for the penalized beta spline, are known and
non-disclosive.
model <- ds.gamlss(
formula = mpg ~ pb(wt), sigma.formula = ~wt,
min.values = min(mtcars$wt),
max.values = max(mtcars$wt),
min.max.names = "wt",
data = "D", family = "NO()"
)
#> Outer iteration 1...
#> Outer iteration 2...
#> Outer iteration 3...However, this might not always be the case and instead, an anonymized minimum and maximum can be used to determine the knots.
model <- ds.gamlss(formula = mpg ~ pb(wt), sigma.formula = ~wt, data = "D", family = "NO()")
#> Outer iteration 1...
#> Outer iteration 2...
#> Outer iteration 3...Note that this implies that the knots are different from a GAMLSS model
that is fit to the whole mctars dataset, and hence the results from
ds.gamlss are slightly different from the GAMLSS model that is fit to
the whole data. However, this should have no impact on the
interpretation of the model.
After the ds.gamlss model has been fit it can be used to predict the
different distribution parameters, e.g. the mu distribution parameter.
Therefore a new data frame with the explanatory variables included in
the model must be created.
newdata <- data.frame(wt = seq(2, 5, by = 0.01))
mu.response <- ds.predict.gamlss(model, newdata, what = "mu", type = "response")
head(mu.response)
#> [,1]
#> 1 27.71961
#> 2 27.64003
#> 3 27.56050
#> 4 27.48102
#> 5 27.40159
#> 6 27.32221- Rigby RA, Stasinopoulos DM. Generalized additive models for location, scale and shape. Journal of the Royal Statistical Society: Series C (Applied Statistics). 2005;54(3):507-54.
- DataSHIELD Developers (2023). dsBaseClient: DataSHIELD Client Functions. R package version 6.3.0.
- Marcon Y (2022). DSLite: ‘DataSHIELD’ Implementation on Local Datasets. R package version 1.4.0, https://CRAN.R-project.org/package=DSLite.