Update Keras to Keras3

.Rbuildignore

@@ -8,3 +8,5 @@
 ^.*\.{jpeg,png,gif,tiff}$
 .cicd-env
 ^LICENSE.md
+^CRAN-SUBMISSION$
+^\.Rcheck$

.gitignore

@@ -7,3 +7,5 @@
 .RData
 *.Rproj*
 .Rproj.user
+*CRAN-SUBMISSION*
+..Rcheck

DESCRIPTION

@@ -1,15 +1,16 @@
 Package: iSubGen
 Title: Integrative Subtype Generation
-Version: 1.0.4
-Date: 2026-01-20
+Version: 1.1.0
+Date: 2026-01-21
 Authors@R: c(
-	person("Mao Tian", role = "cre", email = "[email protected]"),
-	person("Paul Boutros", role = "aut"),
-	person("Natalie Fox", role = "aut"),
-	person("Dan Knight", role = "ctb"))
+    person("Natalie", "Fox", role = "aut"),
+    person("Paul", "Boutros", role = "aut"),
+    person("Mao", "Tian", email = "[email protected]", role = c("cre", "ctb")),
+    person("Dan", "Knight", role = "ctb")
+)
 Description: Multi-data type subtyping, which is data type agnostic and accepts missing data. Subtyping is performed using intermediary assessments created with autoencoders and similarity calculations.
 Depends: R (>= 3.2.3)
-Imports: ConsensusClusterPlus, cluster (>= 1.14.4), keras, tensorflow, philentropy
+Imports: ConsensusClusterPlus, cluster (>= 1.14.4), keras3, tensorflow, philentropy
 Suggests: knitr, rmarkdown
 Remotes:
 	bioc::ConsensusClusterPlus

NAMESPACE

@@ -5,4 +5,4 @@ importFrom("grDevices", "colorRampPalette");
 importFrom("cluster","diana");
 importFrom("ConsensusClusterPlus","ConsensusClusterPlus","calcICL");
 importFrom("philentropy","distance");
-import(tensorflow, keras);
+import(tensorflow, keras3);

NEWS.md

@@ -1,7 +1,8 @@
-# iSubGen 1.0.4 (2026-01-20)
+# iSubGen 1.1.0 (2026-01-21)
 
 ## Changed
 * Update maintainer information and email
+* Update `keras` package to `keras3`
 
 # iSubGen 1.0.3 (2025-06-18)
 

R/create.autoencoder.R

@@ -60,7 +60,7 @@ create.autoencoder <- function(
 		optimizer = 'adam'
 		);
 
-	ae.output.file <- paste0(sub('/$','',model.file.output.dir),'/',data.type,'_model.hdf5');
+	ae.output.file <- paste0(sub('/$','',model.file.output.dir),'/',data.type,'_model.keras');
 	checkpoint <- callback_model_checkpoint(
 		filepath = ae.output.file,
 		save_best_only = TRUE,

R/create.autoencoder.irf.matrix.R

@@ -36,14 +36,14 @@ create.autoencoder.irf.matrix <- function(
 		if (data.type %in% names(autoencoders)) {
 			# load the neural net for the data.type
 			model <- autoencoders[[data.type]];
-			if (is.character(autoencoders[[data.type]]) && grep('hdf5$',autoencoders[[data.type]]) == 1) {
-				model <- load_model_hdf5(
+			if (is.character(autoencoders[[data.type]]) && grep('keras$',autoencoders[[data.type]]) == 1) {
+				model <- load_model(
 					autoencoders[[data.type]],
 					compile = FALSE);
 				}
 
 			# create the autoencoder encoding layers from input to the bottleneck layer
-			intermediate.layer.model <- keras_model(inputs = model$input, outputs = get_layer(model, 'bottleneck')$output);
+			intermediate.layer.model <- keras_model(inputs = model$inputs, outputs = get_layer(model, 'bottleneck')$output);
 
 			# get the bottleneck values from the autoencoder
 			bottleneck.values <- predict(intermediate.layer.model,x = t(data.matrices[[data.type]]));

README.md

@@ -36,6 +36,10 @@ The first step is to load the data. Here the genomic features are rows and patie
 ```{r load-data}
 # Load the library and the data included with the package
 library(iSubGen);
+library(tensorflow);
+library(keras3);
+library(reticulate);
+
 molecular.data <- list();
 for(i in c('cna','methy','snv')) {
   molecular.data[[i]] <- load.molecular.aberration.data(
@@ -72,7 +76,7 @@ A benefit of using autoencoders for compressing features is that autoencoders ca
 ### Create independent reduced features (IRFs)
 
 We create autoencoders individually for each data type.
-The create.autoencoder function is a wrapper function to create an autoencoder using the [keras](https://cran.r-project.org/package=keras) package.
+The create.autoencoder function is a wrapper function to create an autoencoder using the [keras3](https://cran.r-project.org/package=keras3) package.
 Here we create two compressed features for the CNA feature set and one compressed feature for each of the SNV and methylation feature sets.
 
 ```{r ae-matrix}
@@ -231,7 +235,7 @@ IRF: Independent Reduced Features.
 
 iSubGen: integrative Subtype Generation. This tool!
 
-Keras R package: [https://cran.r-project.org/package=keras](https://cran.r-project.org/package=keras)
+keras3 R package: [https://cran.r-project.org/package=keras3](https://cran.r-project.org/package=keras3)
 
 Methylation: a DNA modification where methyl groups of are added to cytosines in DNA to help regulate DNA transcription.
 

man/create.autoencoder.Rd

@@ -1,23 +1,23 @@
 \name{create.autoencoder}
 \alias{create.autoencoder}
 \title{Create an autoencoder for dimensionality reduction}
-\description{Create an autoencoder for dimensionality reduction using keras and tensorflow packages}
+\description{Create an autoencoder for dimensionality reduction using keras3 and tensorflow packages}
 \usage{
 create.autoencoder(data.type, data.matrix, encoder.layers.node.nums = c(15,2),
-autoencoder.activation = 'tanh', optimization.loss.function = 'mean_squared_error', 
+autoencoder.activation = 'tanh', optimization.loss.function = 'mean_squared_error',
 model.file.output.dir = '.')
 }
 \arguments{
   \item{data.type}{data type ID. The ID will be used for naming the output file}
   \item{data.matrix}{matrix with data features as rows and patients as columns}
-  \item{encoder.layers.node.nums}{vector with the number of nodes for each layer when the reducing the feature dimensions within the autoencoder. The autoencoder will be made symmetrically so the number of nodes in each layer will be used in reverse, not repeating the last layer to re encode the features in the autoencoder} 
+  \item{encoder.layers.node.nums}{vector with the number of nodes for each layer when the reducing the feature dimensions within the autoencoder. The autoencoder will be made symmetrically so the number of nodes in each layer will be used in reverse, not repeating the last layer to re encode the features in the autoencoder}
   \item{autoencoder.activation}{activation function to use in the autoencoder}
   \item{optimization.loss.function}{loss function used for optimization while fitting the autoencoder}
   \item{model.file.output.dir}{file location for the autoencoder file}
 }
 \value{
-  \item{autoencoder}{the autoencoder created by the keras package}
-  \item{autoencoder.file}{the hdf5 file that the model was saved in and can be loaded from}
+  \item{autoencoder}{the autoencoder created by the keras3 package}
+  \item{autoencoder.file}{the keras file that the model was saved in and can be loaded from}
 }
 \author{Natalie Fox}
 \examples{

metadata.yaml

@@ -4,5 +4,5 @@ Description: 'Multi-data type subtyping, which is data type agnostic and accepts
 Maintainers: '[email protected]'
 Contributors: ['Natalie Fox', 'Paul C. Boutros', 'Mao Tian']
 Languages: 'R'
-Dependencies: ['ConsensusClusterPlus', 'cluster', 'keras', 'tensorflow', 'philentropy']
+Dependencies: ['ConsensusClusterPlus', 'cluster', 'keras3', 'tensorflow', 'philentropy']
 References: 'Fox, Natalie S., Mao Tian, Alexander L. Markowitz, Syed Haider, Constance H. Li, and Paul C. Boutros. “iSubGen Generates Integrative Disease Subtypes by Pairwise Similarity Assessment.” Cell Reports Methods 4, no. 11 (2024): 100884. https://doi.org/10.1016/j.crmeth.2024.100884.'

vignettes/iSubGenGuide.Rnw

@@ -91,6 +91,9 @@ The first step is to load the data. Here the genomic features are rows and patie
 <<load-data, eval = FALSE>>=
 # Load the library and the data included with the package
 library(iSubGen);
+library(tensorflow);
+library(keras3);
+library(reticulate);
 molecular.data <- list();
 for (i in c('cna','methy','snv')) {
   molecular.data[[i]] <- load.molecular.aberration.data(
@@ -128,7 +131,7 @@ A benefit of using autoencoders for compressing features is that autoencoders ca
 \subsection{Create independent reduced features (IRFs)}
 
 We create autoencoders individually for each data type.
-The create.autoencoder function is a wrapper function to create an autoencoder using the \href{https://cran.r-project.org/package=keras}{keras} package.
+The create.autoencoder function is a wrapper function to create an autoencoder using the \href{https://cran.r-project.org/package=keras3}{keras3} package.
 Here we create two compressed features for the CNA feature set and one compressed feature for each of the SNV and methylation feature sets.
 
 <<ae-matrix, eval = FALSE>>=
@@ -276,7 +279,7 @@ DNA segments can be deleted so there are less than 2 copies. \\
 
 \noindent \textbf{iSubGen:} integrative Subtype Generation. This tool!\\
 
-\noindent \textbf{Keras R package:} \url{https://cran.r-project.org/package=keras} \\
+\noindent \textbf{keras3 R package:} \url{https://cran.r-project.org/package=keras3} \\
 
 \noindent \textbf{Methylation:} a DNA modification where methyl groups of are added to cytosines in DNA to help regulate DNA transcription. \\