AlphaTherapy

Introduction

AlphaTherapy is a theoretical proof-of-concept AI framework for the rational design of sequential drug treatments for tumors across diverse drugs and tumor types.

Users can create a scenario by customizing input parameters tailored to their specific problem related to sequential therapy. Specifically, AlphaTherapy is designed to investigate sequential drug treatments applicable for different scenarios, including for a specified cell line (parameter: cell line), for a specified treatment duration (parameter: terminal step) and for a constrained sequential drug combination space (parameter: drug pool).

Table of contents

• AlphaTherapy
  • Introduction
  • Table of contents
  • Requirements
  • Installation from github
  • Installation from Docker image
  • Usage
    • 1. Rational design of sequential drug treatments for tumors by AlphaTherapy
      • 1.1 Drug pool construction
      • 1.2 Write configurations
      • 1.3 Run RL agents
      • 1.4 Merge results
    • 2. Downstream analysis
  • Citation
  • Contacts

Requirements

• Python == 3.6.13
• torch == 1.10.2
• gym == 0.15.4
• tensorboard == 2.2.2
• rdkit == 2021.03.5
• openpyxl, numpy, pandas, scipy, matplotlib, scikit-learn Python modules are also needed
• R == 4.3.1
• clusterProfiler, org.Hs.eg.db, usefun R packages are also needed

Installation from Docker image

docker pull xiaohanchen/alphatherapy:v1

Installation from github

1. Downloading AlphaTherapy

git clone https://github.com/bm2-lab/AlphaTherapy.git

2. Installing some packages

cd ./scripts/gym_cell_model
pip install -e .
cd ../tianshou-master
pip install -e .

Usage

1. Rational design of sequential drug treatments for tumors by AlphaTherapy

This section will generate a list of effective sequential drug combinations for user-specified scenarios, sorted by episode reward. For illustration, we use a specific scenario with the following three key parameters:

1. Drug Pool: A drug set containing 395 drugs (387 anticancer drugs approved by FDA and 8 other drugs).
2. Cell Line: MCF7, a commonly used breast cancer cell line.
3. Termination Step: Step 2, indicating the drug duration time. One step indicates 24 hour drug administration.

1.1 Drug pool construction

python ./scripts/utils/build_drugset.py --input FDA_drugset.xlsx --output FDA_drugset.pickle

The command facilitates the construction of a drug set by processing an input Excel file and outputing a pickle file for use in cell model environments.

Parameters

• input: the file name of an Excel file containing the compound information of the drug set. If users wish to create a custom drug set, they must store thier input file in ./data/ and ensure that the input Excel file contains the following five required columns.
  1. Index: The index of the compound.
  2. Name: The name of the compound.
  3. Target: The target(s) of the compound. Multiple targets should be separated by commas. If the target is unknown, use "Unknown".
  4. Pathway: The pathway associated with the compound. If the pathway is unknown, use "Unknown".
  5. SMILES: The SMILES (Simplified Molecular Input Line Entry System) notation of the compound. Ensure that the correct SMILES notation is provided.
• output: The output is the file name of a pickle file containing the processed compound information of the drug set in ./scripts/gym_cell_model/gym_cell_model/data directory.

1.2 Write configurations

python ./scripts/utils/write_scenario_configs.py --drugset_file_name FDA_drugset.pickle --drugset_short_name FDA --cell_line MCF7 --terminal_step 2

This command prepares configurations prior to running the RL agents by modifying two config files:

• env_cpd.config: This file contains configurations for the cell model environment located in /scripts/gym_cell_model/gym_cell_model/config/. The command will append configurations named in the format ENV_[drugset_short_name]_[cell_line]_[terminal_step], like ENV_FDA_MCF7_2.
• RL_agent.config: This file contains configurations for the RL agents located in /scripts/AlphaTherapy/config/. The command will append configurations named in the format ENV_[drugset_short_name]_[cell_line]_[terminal_step]_SEED[i], where i ranges from 1 to 10, like ENV_FDA_MCF7_2_SEED1.

Parameters

• drugset_file_name: the pickle file name of drug set which saved in the above command.
• drugset_short_name: a phrase or a short name for easy reference of the drug set.
• cell_line: the cell line you want to research.
• terminal_step: a number of the terminal step you want to reearch.

Notably, we only add the information of the scenario to RL_agents.config. You also can modify the model training parameters in RL_agents.config.

1.3 Run RL agents

nohup python ./scripts/AlphaTherapy/model/train_RL_agents.py --drugset_short_name FDA --cell_line MCF7 --terminal_step 2 --max_workers 10 &

This command trains RL agents through interaction with the cell model. It will generate policy.pth, plot_train_data.csv and plot_test_data.csv files within new directories like "./scripts/AlphaTherapy/working_log/ENV_FDA_MCF7_STEP2_SEED1". These files serve as records for model training, capturing all generated .pth files during runtime and the reward changes during training process.

Parameters

• drugset_short_name, cell_line, terminal_step is the same as above.
• max_workers: maximum number of available CPUs.

1.4 Merge results

python ./scripts/AlphaTherapy/model/output_result.py --drugset_short_name FDA --cell_line MCF7 --terminal_step 2

This command will organize the sequential drug combination results for each scenario, like ENV_FDA_A375_STEP2, and finally output a file named ENV_FDA_A375_STEP2.csv, stored in /output/AlphaTherapy/.

Parameters

• drugset_short_name, cell_line, terminal_step is the same as above.

2. Downstream analysis

python ./scripts/downstream_analysis/src/run_downstram_analysis.py --combos_file_name ENV_FDA_MCF7_STEP2.csv --env_name ENV_FDA_MCF7_STEP2 --drugA_name "Morin Hydrate" --drugB_name "Afatinib (BIBW2992)"

This command will run downstream analysis and save the results in file like in /output/downstream_analysis/.

Parameters

• combos_file_name: the name of file generated in the previous command like ENV_FDA_MCF7_STEP2.csv.
• env_name: the config name of cell model, like ENV_FDA_A375_STEP2.
• drugA_name: the name of the first frug from drug combos in the combos_file_name.csv.
• drugB_name: the name of the second frug from drug combos in the combos_file_name.csv.

Reproducity

To facilitate reliable reproduction of the results presented in the paper, we have provided a folder named reproducibility.

1. Running main.sh script in the main project directory that generates all the results required for plotting.
2. The scripts within the notebooks directory, located under the reproductity folder, enable users to recreate the figures presented in the manuscript.

Citation

Contacts

[email protected]

[email protected]