| Type: | Package |
| Title: | Optimal Decision Trees Algorithm |
| Version: | 1.0.0 |
| Maintainer: | Katyna Sada Del Real <ksada@unav.es> |
| Description: | Implements a tree-based method specifically designed for personalized medicine applications. By using genomic and mutational data, 'ODT' efficiently identifies optimal drug recommendations tailored to individual patient profiles. The 'ODT' algorithm constructs decision trees that bifurcate at each node, selecting the most relevant markers (discrete or continuous) and corresponding treatments, thus ensuring that recommendations are both personalized and statistically robust. This iterative approach enhances therapeutic decision-making by refining treatment suggestions until a predefined group size is achieved. Moreover, the simplicity and interpretability of the resulting trees make the method accessible to healthcare professionals. Includes functions for training the decision tree, making predictions on new samples or patients, and visualizing the resulting tree. For detailed insights into the methodology, please refer to Gimeno et al. (2023) <doi:10.1093/bib/bbad200>. |
| Depends: | R (≥ 4.0), matrixStats, partykit, data.tree, stats |
| Imports: | magick, DiagrammeRsvg, grDevices, DiagrammeR, rsvg |
| Suggests: | RUnit, Matrix, rmarkdown, robustbase, knitr |
| License: | Artistic-2.0 |
| LazyData: | true |
| LazyDataCompression: | xz |
| RoxygenNote: | 7.3.2 |
| Encoding: | UTF-8 |
| VignetteBuilder: | knitr |
| NeedsCompilation: | no |
| Packaged: | 2024-10-17 10:19:25 UTC; katyna |
| Repository: | CRAN |
| Date/Publication: | 2024-10-18 10:50:43 UTC |
| Author: | Maddi Eceiza [aut], Lucia Ruiz [aut], Angel Rubio [aut], Katyna Sada Del Real [aut, cre] |
ODT Internal Functions
Description
Internal functions used by 'trainTree' in the various steps of the algorithm. These functions are not intended for direct use by end users and are primarily for internal logic and calculations within the package.
This internal function calculates the optimal treatment based on patient sensitivity data.
This internal function calculates the optimal split point based on a specified gene in the expression data.
This internal function calculates the summed IC50 value based on the response of patients to a specific gene.
This internal function identifies the optimal split point in expression data based on patient sensitivity (IC50).
This internal function recursively builds a decision tree based on expression data and patient sensitivity (IC50).
This internal function identifies the optimal split point in a mutation matrix based on patient sensitivity data (IC50) and the presence of specific mutations.
This internal function recursively builds a decision tree based on patient data and drug sensitivity responses, specifically designed for mutation data.
Usage
getTreatment(PatientSensitivity, weights)
getSplit(gene, PatientData, PatientSensitivity)
getsumic50v3(genePatientResponse, PatientSensitivity)
findsplitExp(
PatientSensitivity,
PatientData,
minimum = 1,
weights = NULL,
verbose = FALSE
)
growtreeExp(
id = 1L,
PatientSensitivity,
PatientData,
minbucket = 10,
weights = NULL
)
findsplitMut(PatientSensitivity, X, minimum = 1, weights)
growtreeMut(
id = 1L,
PatientSensitivity,
PatientData,
minbucket = 10,
weights = NULL,
findsplit = findsplitMut
)
Arguments
PatientSensitivity |
A matrix representing drug response values (e.g., IC50), where rows correspond to patients/samples and columns correspond to drugs. |
weights |
A numeric vector indicating which samples are under study; defaults to NULL, meaning all samples are considered. |
gene |
The index of the gene used for splitting the data. |
PatientData |
A matrix representing patient features, where rows correspond to patients/samples and columns correspond to genes/features. |
genePatientResponse |
A numeric vector representing the response of patients to a particular gene. |
minimum |
An integer specifying the minimum number of samples required for a valid split (default is 1). |
verbose |
A logical value indicating whether to print additional information during execution (default is FALSE). |
id |
A unique identifier for the node in the decision tree (default is 1L). |
minbucket |
An integer specifying the minimum number of samples required in a child node for further splitting (default is 10). |
X |
(PatientData) A matrix representing patient features, where rows correspond to patients/samples and columns correspond to genes/features. |
findsplit |
A function used to find the best split; defaults to 'findsplitMut'. |
Value
A list of internal outputs generated by these functions.
An integer indicating the index of the optimal treatment based on the minimum sum of sensitivity values.
A list containing: - 'sumic50': The minimum summed IC50 values for the two groups. - 'Treatment1': The treatment associated with the first group. - 'Treatment2': The treatment associated with the second group. - 'split': The index of the optimal split point. - 'expressionSplit': The expression value at the split point adjusted by a small epsilon.
A numeric value representing the minimum summed IC50 value derived from the patient responses.
A 'partysplit' object representing the optimal split based on the specified patient sensitivity and expression data.
A 'partynode' object representing the current node and its children in the decision tree.
A 'partysplit' object if a valid split is found, or 'NULL' if no valid split can be determined.
A 'partynode' object representing the current node and its children in the decision tree.
drug_response_w12 data
Description
A matrix containing drug response values (IC50 values) from patients retrieved from waves 1 and 2. Used as a toy example in trainTree and predictTree and niceTree
Usage
data("drug_response_w12")Format
The format is: num [1:247, 1:119] 2.710983 2.8755433 3.4390103 2.6527257...
Examples
data(drug_response_w12)
drug_response_w34 data
Description
A matrix containing drug response values (IC50 values) from patients retrieved from waves 3 and 4. Used as a toy example in trainTree and predictTree and niceTree
Usage
data("drug_response_w34")Format
The format is: num [1:247, 1:119] 3.4156359 3.2345985 3.1836058 3.7874252...
Examples
data(drug_response_w34)
expression_w12 Data Set
Description
A dataframe containing gene expression values (with different types of genes) from patients retrieved from waves 1 and 2. Used as a toy example in trainTree and predictTree and niceTree
Usage
data("expression_w12")Format
A data frame with 22843 observations on the following 247 variables, where there are 247 patients from the waves and 22843 different genes are considered.
The format is: 22843 obs. 247 variables
Examples
data(expression_w12)
expression_w34 Data Set
Description
A dataframe containing gene expression values (with different types of genes) from patients retrieved from waves 3 and 4. Used as a toy example in trainTree and predictTree and niceTree
Usage
data("expression_w34")Format
A data frame with 22843 observations on the following 142 variables, where there are 142 patients from the waves and 22843 different genes are considered.
The format is: 22843 obs. 142 variables
Examples
data(expression_w34)
mutations_w12 Data Set
Description
A binary matrix containing mutation values, wether the mutation is present or not in the patient from patients retrieved from waves 1 and 2. Used as a toy example in trainTree and predictTree and niceTree
Usage
data("mutations_w12")Format
A binary matrix with,where there are 247 patients from the waves and 70 different mutations are considered.
The format is: num [1:247, 1:70] 0 0 0 0 0 1...
Examples
data(mutations_w12)
mutations_w34 Data Set
Description
A binary matrix containing mutation values, wether the mutation is present or not in the patient from patients retrieved from waves 3 and 4. Used as a toy example in trainTree and predictTree and niceTree
Usage
data("mutations_w34")Format
A binary matrix with,where there are 142 patients from the waves and 70 different mutations are considered.
The format is: num [1:142, 1:70] 0 0 0 1 0 0...
Examples
data(mutations_w34)
niceTree function
Description
A graphical display of the tree. It can also be saved as an image in the selected directory.
Usage
niceTree(
tree,
folder = NULL,
colors = c("", "#367592", "#39A7AE", "#96D6B6", "#FDE5B0", "#F3908B", "#E36192",
"#8E4884", "#A83333"),
fontname = "Roboto",
fontstyle = "plain",
shape = "diamond",
output_format = "png"
)
Arguments
tree |
A party of the trained tree with the treatments assigned to each node. |
folder |
Directory to save the image (default is the current working directory). |
colors |
A vector of colors for the boxes. Can include hex color codes (e.g., "#FFFFFF"). |
fontname |
The name of the font to use for the text labels (default is "Roboto"). |
fontstyle |
The style of the font (e.g., "plain", "italic", "bold"). |
shape |
The format of the boxes for the different genes (e.g., "diamond", "box"). |
output_format |
The image format for saving (e.g., "png", "jpg", "svg", "pdf"). |
Details
The user has already defined a style for the plot; the parameters are set if not modified when calling niceTree.
Value
(Invisibly) returns a list. The representation of the tree in the command window and the plot of the tree.
Examples
# Basic example of how to perform niceTree:
data("mutations_w12")
data("drug_response_w12")
ODTmut <- trainTree(PatientData = mutations_w12,
PatientSensitivity = drug_response_w12, minbucket = 10)
niceTree(ODTmut)
# Example for plotting the tree trained for gene expressions:
data("expression_w34")
data("drug_response_w34")
ODTExp <- trainTree(PatientData = expression_w34,
PatientSensitivity = drug_response_w34, minbucket = 20)
niceTree(ODTExp)
Predict Treatment Outcomes with a Trained Decision Tree
Description
This function utilizes a trained decision tree model (ODT) to predict treatment outcomes for test data based on patient sensitivity data and features, such as mutations or gene expression profiles.
Usage
predictTree(tree, PatientData, PatientSensitivityTrain)
Arguments
tree |
A trained decision tree object created by the 'trainTree' function. |
PatientData |
A matrix representing patient features, where rows correspond to patients/samples and columns correspond to genes/features. This matrix can contain:
|
PatientSensitivityTrain |
A matrix containing the drug response values of the **training dataset**. In this matrix, rows correspond to patients, and columns correspond to drugs. This matrix is used solely for extracting treatment names and is not used in the prediction process itself. |
Value
A factor representing the assigned treatment for each node in the decision tree based on the provided patient data and sensitivity.
Examples
# Example 1: Prediction using mutation data
data("mutations_w12")
data("mutations_w34")
data("drug_response_w12")
ODTmut <- trainTree(PatientData = mutations_w12,
PatientSensitivity = drug_response_w12,
minbucket = 10)
ODTmut
ODT_mutpred <- predictTree(tree = ODTmut,
PatientSensitivityTrain = drug_response_w12,
PatientData = mutations_w34)
# Example 2: Prediction using gene expression data
data("expression_w34")
data("expression_w12")
data("drug_response_w34")
ODTExp <- trainTree(PatientData = expression_w34,
PatientSensitivity = drug_response_w34,
minbucket = 20)
ODTExp
ODT_EXPpred <- predictTree(tree = ODTExp,
PatientSensitivityTrain = drug_response_w34,
PatientData = expression_w12)
trainTree Function
Description
This function trains a decision tree model based on patient data, which can either be gene expression levels or a binary matrix indicating mutations.
Usage
trainTree(PatientData, PatientSensitivity, minbucket = 20)
Arguments
PatientData |
A matrix representing patient features, where rows correspond to patients/samples and columns correspond to genes/features. This matrix can contain:
|
PatientSensitivity |
A matrix representing drug response values, where rows correspond to patients in the same order as in 'PatientData', and columns correspond to drugs. Higher values indicate greater drug resistance and, consequently, lower sensitivity to treatment. This matrix can represent various measures of drug response, such as IC50 values or area under the drug response curve (AUC). Depending on the interpretation of these values, users may need to adjust the sign of this data. |
minbucket |
An integer specifying the minimum number of patients required in a node to allow for a split. |
Value
An object of class 'party' representing the trained decision tree, with the assigned treatments for each node.
Examples
# Basic example of using the trainTree function with mutational data
data("drug_response_w12")
data("mutations_w12")
ODTmut <- trainTree(PatientData = mutations_w12,
PatientSensitivity = drug_response_w12,
minbucket = 10)
plot(ODTmut)
# Example using gene expression data instead
data("drug_response_w34")
data("expression_w34")
ODTExp <- trainTree(PatientData = expression_w34,
PatientSensitivity = drug_response_w34,
minbucket = 20)
plot(ODTExp)