Type: Package
Title: Optimal Dose Escalation Using Deep Reinforcement Learning
Version: 1.0.3
Description: An implementation to compute an optimal dose escalation rule using deep reinforcement learning in phase I oncology trials (Matsuura et al. (2023) <doi:10.1080/10543406.2023.2170402>). The dose escalation rule can directly optimize the percentages of correct selection (PCS) of the maximum tolerated dose (MTD).
URL: https://github.com/MatsuuraKentaro/RLescalation
BugReports: https://github.com/MatsuuraKentaro/RLescalation/issues
VignetteBuilder: knitr
License: MIT + file LICENSE
Encoding: UTF-8
Language: en-US
RoxygenNote: 7.3.2
Imports: glue, R6, nleqslv, reticulate, stats, utils, zip
Suggests: knitr, rmarkdown
Collate: 'timer.R' 'train_algo.R' 'utils.R' 'escalation_rule.R' 'rl_dnn_config.R' 'rl_config_set.R' 'compute_rl_scenarios.R' 'learn_escalation_rule.R' 'setup_python.R' 'zzz.R' 'simulate_one_trial.R'
NeedsCompilation: no
Packaged: 2025-10-07 13:26:45 UTC; kmatsuu
Author: Kentaro Matsuura ORCID iD [aut, cre, cph]
Maintainer: Kentaro Matsuura <matsuurakentaro55@gmail.com>
Repository: CRAN
Date/Publication: 2025-10-07 13:50:07 UTC

EscalationRule Class

Description

This class represents an escalation rule that generates a next escalation.

Public fields

policy

The RLlib policy that is a Python object.

dir

Directory path of the escalation rule (policy).

dirpath

Full path to the directory of the escalation rule.

created_at

Created time of this object.

info

Information when learning the escalation rule.

input

Inputs for learning the escalation rule.

log

The log of scores during the learning of the escalation rule.

checkpoints

The integer vector of iteration counts for checkpoints.

checkpoints_paths

The paths to the directories where each checkpoint is stored.

Methods

Public methods

Method new()

Create a new EscalationRule object.

Usage
EscalationRule$new(dir = "latest", base_dir = "escalation_rules")
Arguments
dir

A character value. A directory name or path where an escalation rule is outputted. By default, the latest escalation rule is searched in 'base_dir'.

base_dir

A character value. A directory path that is used as the parent directory if the 'dir' argument is a directory name and is not used otherwise.

Method opt_action()

Compute optimal action probabilities using the obtained escalation rule for data of N and DLT.

Usage
EscalationRule$opt_action(current_dose, data_Ns, data_DLTs)
Arguments
current_dose

An integer value. This is the current dose index, which is within 1:J.

data_Ns

A numeric vector. The cumulative number of patients assigned to each dose in your clinical trial.

data_DLTs

A numeric vector. The cumulative number of DLTs corresponding to each dose for the 'data_Ns' argument.

Returns

A character that represents the optimal action. One of the followings: down, stay, up, MTD_1, ..., MTD_J, no_MTD

Method resume_learning()

Resume learning the escalation rule. This function updates the original EscalationRule object.

Usage
EscalationRule$resume_learning(iter)
Arguments
iter

A number of additional iterations.

Returns

An updated EscalationRule object.

Method set_info()

Set information when learning the escalation rule.

Usage
EscalationRule$set_info(info, input, log, checkpoints)
Arguments
info

Information when learning the escalation rule.

input

Inputs for learning the escalation rule.

log

The log of scores during the learning of the escalation rule.

checkpoints

The paths to the directories where each checkpoint is stored.

Method print()

Print function for EscalationRule object

Usage
EscalationRule$print()

Method clone()

The objects of this class are cloneable with this method.

Usage
EscalationRule$clone(deep = FALSE)
Arguments
deep

Whether to make a deep clone.

Clean the Python Virtual Environment

Description

Clean the Python Virtual Environment

Usage

clean_python_settings(envname = "r-RLescalation")

Arguments

envname

Python virtual environment name.

Compute DLT Probability Scenarios for Reinforcement Learning

Description

Compute the scenarios described in Sect. 2.2 of the original paper.

Usage

compute_rl_scenarios(J, target, epsilon, delta, lower = 0.1, upper = 0.8)

Arguments

J

A positive integer value. The number of doses.

target

A positive numeric value. The target DLT probability.

epsilon

A positive numeric value. The acceptable range of target DLT probabilities is defined as [target - epsilon, target + epsilon].

delta

A positive numeric value. The unacceptable ranges of target DLT probabilities are defined as [0, target - delta] and [target + delta, 1].

lower

A positive numeric value. Values lower than lower are clipped. Default is 0.1, which is modified from Sect. 2.2 of the original paper.

upper

A positive numeric value. Values higher than upper are clipped. Default is 0.8.

Value

A named list of three elements: - prob: a list of DLT probability scenarios - MTD: a list of true MTD indices (Note that -1 means "no MTD") - weight: a vector of weights for each scenario

Examples

scenarios <- compute_rl_scenarios(J = 6, target = 0.25, epsilon = 0.04, delta = 0.1)
print(scenarios)

Build an Optimal Dose Escalation Rule using Reinforcement Learning

Description

Build an Optimal Dose Escalation Rule using Reinforcement Learning

Usage

learn_escalation_rule(
  J,
  target,
  epsilon,
  delta,
  N_total,
  N_cohort,
  seed = NULL,
  rl_config = rl_config_set(),
  rl_scenarios = NULL,
  output_dir = format(Sys.time(), "%Y%m%d_%H%M%S"),
  output_base_dir = "escalation_rules",
  checkpoint_dir = "checkpoints"
)

Arguments

J

A positive integer value. The number of doses.

target

A positive numeric value. The target DLT probability.

epsilon

A positive numeric value. The acceptable range of target DLT probabilities is defined as [target - epsilon, target + epsilon].

delta

A positive numeric value. The unacceptable ranges of target DLT probabilities are defined as [0, target - delta] and [target + delta, 1].

N_total

A positive integer value. The total number of patients.

N_cohort

A positive integer value. The number of patients for each cohort.

seed

An integer value. Random seed for reinforcement learning.

rl_config

A list. Other settings for reinforcement learning. See rl_config_set for details.

rl_scenarios

A list. Scenarios used for reinforcement learning. Default is NULL (use scenarios in the Sect. 2.2 of the original paper). See compute_rl_scenarios for details.

output_dir

A character value. Directory name or path to store the built escalation rule. Default is the current datetime.

output_base_dir

A character value. Parent directory path where the built escalation rule will be stored. Valid only if 'output_dir' does not contain '/'. Default is "escalation_rules".

checkpoint_dir

A character value. Parent directory path to save checkpoints. It enables you to resume learning from that point onwards. Default is "checkpoints".

Value

An EscalationRule object.

Examples

library(RLescalation)

# We obtain an optimal dose escalation rule by executing `learn_escalation_rule()`.
## Not run: 
escalation_rule <- learn_escalation_rule(
  J = 6, target = 0.25, epsilon = 0.04, delta = 0.1,
  N_total = 36, N_cohort = 3, seed = 123,
  rl_config = rl_config_set(iter = 1000)
)
## End(Not run)

Configuration of Reinforcement Learning

Description

Mainly settings for the arguments of the training() function. Not compatible with the new API stack introduced in Ray 2.10.0.

Usage

rl_config_set(
  iter = 1000L,
  save_start_iter = NULL,
  save_every_iter = NULL,
  cores = 4L,
  gamma = 1,
  lr = 5e-05,
  train_batch_size = 10000L,
  model = rl_dnn_config(),
  sgd_minibatch_size = 200L,
  num_sgd_iter = 20L,
  ...
)

Arguments

iter

A positive integer value. Number of iterations.

save_start_iter, save_every_iter

An integer value. Save checkpoints every 'save_every_iter' iterations starting from 'save_start_iter' or later.

cores

A positive integer value. Number of CPU cores used for learning.

gamma

A positive numeric value. Discount factor of the Markov decision process. Default is 1.0 (not discount).

lr

A positive numeric value. Learning rate (default 5e-5). You can set a learning schedule instead of a learning rate.

train_batch_size

A positive integer value. Training batch size. Deprecated on the new API stack.

model

A list. Arguments passed into the policy model. See rl_dnn_config for details.

sgd_minibatch_size

A positive integer value. Total SGD batch size across all devices for SGD. Deprecated on the new API stack.

num_sgd_iter

A positive integer value. Number of SGD iterations in each outer loop.

...

Other settings for training(). See the arguments of the training() function in the source code of RLlib. https://github.com/ray-project/ray/blob/master/rllib/algorithms/algorithm_config.py https://github.com/ray-project/ray/blob/master/rllib/algorithms/ppo/ppo.py

Value

A list of reinforcement learning configuration parameters

Examples

## Not run: 
escalation_rule <- learn_escalation_rule(
  J = 6, target = 0.25, epsilon = 0.04, delta = 0.1,
  N_total = 36, N_cohort = 3, seed = 123,
  # We change `iter` to 200 and `cores` for reinforcement learning to 2
  rl_config = rl_config_set(iter = 200, cores = 2)
)
## End(Not run)

DNN Configuration for Reinforcement Learning

Description

DNN (deep neural network) configuration for reinforcement learning. For detail, see Section 3.1 of the original paper.

Usage

rl_dnn_config(
  fcnet_hiddens = c(256L, 256L),
  fcnet_activation = c("relu", "tanh", "swish", "silu", "linear"),
  ...
)

Arguments

fcnet_hiddens

A positive integer vector. Numbers of units of the intermediate layers.

fcnet_activation

A character value specifying the activation function. Possible values are "ReLU" (default), "tanh", "Swish" (or "SiLU"), or "linear".

...

Other configurations. See source code of RLlib. https://github.com/ray-project/ray/blob/master/rllib/models/catalog.py

Value

A list of DNN configuration parameters

Examples

## Not run: 
escalation_rule <- learn_escalation_rule(
  J = 6, target = 0.25, epsilon = 0.04, delta = 0.1,
  N_total = 36, N_cohort = 3, seed = 123,
  rl_config = rl_config_set(
    iter = 1000, 
    # We change the DNN model
    model = rl_dnn_config(fcnet_hiddens = c(512L, 512L), fcnet_activation = "tanh")
  )
)
## End(Not run)

Setting up a Python Virtual Environment

Description

Setting up a Python virtual environment for the Ray package, which includes the RLlib library for reinforcement learning.

Usage

setup_python(envname = "r-RLescalation")

Arguments

envname

Python virtual environment name.

Simulate One Trial Using an Obtained Optimal Dose Escalation Rule

Description

Simulate One Trial Using an Obtained Optimal Dose Escalation Rule

Usage

simulate_one_trial(escalation_rule, prob_true, seed = NULL)

Arguments

escalation_rule

An object of class EscalationRule specifying an obtained optimal dose escalation rule.

prob_true

A numeric vector specifying the true DLT probabilities.

seed

An integer value. Random seed for data generation in this trial.

Value

A data frame which contains the cohort ID, the assigned dose, the number of assigned patients, the number of DLTs, and the recommended action including down, stay, up, MTD_1, ..., MTD_J, no_MTD, and fail to determine MTD.

Examples

library(RLescalation)

## Not run: 
escalation_rule <- learn_escalation_rule(
  J = 6, target = 0.25, epsilon = 0.04, delta = 0.1,
  N_total = 36, N_cohort = 3, seed = 123,
  rl_config = rl_config_set(iter = 1000)
)
## End(Not run)

prob_true <- c(0.03, 0.13, 0.17, 0.19, 0.26, 0.31)

# Simulate one trial using the obtained `escalation_rule`
## Not run: 
sim_one <- simulate_one_trial(escalation_rule, prob_true, seed = 123)
## End(Not run)