Replication Package for article Harnessing Machine Learning Models to Repurpose Drugs Targeting HIV-1 Integrase, Protease, and Reverse Transcriptase

# Replication Package for paper "Harnessing Machine Learning Models to Repurpose Drugs Targeting HIV-1 Integrase, Protease, and Reverse Transcriptase" This package reproduces the experiments described in the paper:Harnessing Machine Learning Models to Repurpose Drugs Targeting HIV-1 Integrase, Protease, and Reverse Transcriptase Ciprian-Bogdan Chirila, University Politehnica Timisoara, 2 V. Parvan Avenue, Timisoara 300223, Romania, [e-mail] Luminita Crisan, "Coriolan Dragulescu" Institute of Chemistry, 24 M. Viteazu Avenue,Timisoara 300223, Romania, [e-mail] Keywords: antiviral, drug repurposing, machine learning # Content The repository is organized into the following folders: code – contains the Python source code of the prototype.pred-dc – contains the predictions generated on the DrugCentral dataset by the first level of the prototype.pred-dc-ml – contains the predictions on the DrugCentral dataset produced by the second level of the prototype, which is based on a meta-learner.source dataset – contains the original datasets downloaded from the ChEMBL and DrugCentral sources.source of figures – contains the original files used to generate the figures included in the manuscript.train-ch – contains the training data derived from the ChEMBL dataset for the first level of the prototype.train-ch-ml – contains the training data generated by the second level of the prototype, based on the meta-learner. # Requirements For CPU processing- Python 3.10.10- TensorFlow 2.10.0- Tested on Windows 10, Windows 11 For GPU processing:- Python 3.10.10- Nvidia Cuda 11.8.0- cudnn 8.6.0.163- tensorflow-2.10.0-cp310-cp310-win_amd64.whl- Tested on Windows 10, Windows 11 # Installation In the root folder of the uncompressed archive you have to run the following commands: # For CPU processingpip install tensorflow-cpu==2.10.0pip install -r requirements.txt # For GPU processing:pip install tensorflow-2.10.0-cp310-cp310-win_amd64.whlpip install -r requirements.txt # Running the Experiment Step 1: Training the Base Models In the root folder of the uncompressed archive you have to run the following commands: run-l1.bat- Trains the 10 base models and generates first level predictions.- The training is made on 9 use case files coresponding to the 3 enzymes x 3 descriptors.- The models train on the data from ChEMBL database located in ./train-ch/data/ folder. - The trained models predict the inputs from DrugCentral database located in ./pred-dc/data/ folder. python ./code/models/metric-performance.py- Computes the performance metrics. python ./code/plot-performance-metrics.py - Compiles the tables with the performance metrics data. python ./code/plot-met-dist-histplot.py - Plots the performance metrics distribution histogram. # Output Step 1 - The predictions from ChEMBL database are located in ./train-ch/predictions/ folder.- The predictions from DrugCentral database are located in ./pred-dc/predictions/ folder.- The performance metrics from ChEMBL are located in ./train-ch/metrics/ folder.- The performance metrics plots from ChEMBL are located in ./train-ch/metrics/metrics-plots/ folder. # Running the Experiment Step 2 Training the Meta-Learner run-l2.bat- Trains the metalearner on 9578 base level model predictions with varying hyperparameters.- The models train on the input data processed from the ChEMBL predictions located in ./train-ch-ml/data/ folder. - The trained models predict the input data processed from DrugCentral predictions located in ./pred-dc-ml/data/ folder. # Output Step 2 - The predictions of ChEMBL database are located in ./train-ch-ml/predictions/ folder.- The predictions of DrugCentral database are located in ./pred-dc-ml/predictions/ folder. # Notes If a GPU is available, TensorFlow 2.10 will use it automatically.If not, it will fall back to CPU execution.The code was tested using both TensorFlow CPU and GPU builds.The run-all.bat script runs both steps in a single command.# 论文《利用机器学习模型对靶向HIV-1整合酶、蛋白酶与逆转录酶的药物进行重定位》复现包 本复现包用于复现论文《利用机器学习模型对靶向HIV-1整合酶、蛋白酶与逆转录酶的药物进行重定位》中所述的实验。 作者： Ciprian-Bogdan Chirila，蒂米什瓦拉理工大学，罗马尼亚蒂米什瓦拉V. Parvan大道2号，邮编300223，邮箱：[e-mail] Luminita Crisan，“科里奥兰·德拉古列斯库”化学研究所，罗马尼亚蒂米什瓦拉M. Viteazu大道24号，邮编300223，邮箱：[e-mail] 关键词：抗病毒，药物重定位，机器学习 ## 内容 本仓库按以下目录组织： - code：包含原型系统的Python源代码。 - pred-dc：包含原型系统第一层级在DrugCentral数据库（DrugCentral）上生成的预测结果。 - pred-dc-ml：包含原型系统第二层级（基于元学习器（meta-learner））在DrugCentral数据库上生成的预测结果。 - source dataset：包含从ChEMBL数据库（ChEMBL）与DrugCentral数据库下载的原始数据集。 - source of figures：包含用于生成论文手稿中插图的原始文件。 - train-ch：包含原型系统第一层级所用的、源自ChEMBL数据库的训练数据。 - train-ch-ml：包含原型系统第二层级（基于元学习器）生成的训练数据。 ## 运行环境要求 ### CPU运行环境 - Python 3.10.10 - TensorFlow 2.10.0 - 已在Windows 10、Windows 11系统上测试 ### GPU运行环境 - Python 3.10.10 - Nvidia Cuda 11.8.0 - cudnn 8.6.0.163 - tensorflow-2.10.0-cp310-cp310-win_amd64.whl - 已在Windows 10、Windows 11系统上测试 ## 安装步骤 在解压后的归档文件根目录下执行以下命令： ### CPU运行环境安装 pip install tensorflow-cpu==2.10.0 pip install -r requirements.txt ### GPU运行环境安装 pip install tensorflow-2.10.0-cp310-cp310-win_amd64.whl pip install -r requirements.txt ## 实验步骤1：训练基础模型（base models） 在解压后的归档文件根目录下执行以下命令： 1. `run-l1.bat`：训练10个基础模型并生成第一层级预测结果。本次训练基于对应3种酶×3种描述符的9个用例文件完成，模型以`./train-ch/data/`目录下的ChEMBL数据库数据作为训练集，对`./pred-dc/data/`目录下的DrugCentral数据库输入数据进行预测。 2. `python ./code/models/metric-performance.py`：计算性能指标。 3. `python ./code/plot-performance-metrics.py`：生成包含性能指标数据的表格。 4. `python ./code/plot-met-dist-histplot.py`：绘制性能指标分布直方图。 ## 步骤1输出结果 - ChEMBL数据库的预测结果存储于`./train-ch/predictions/`目录。 - DrugCentral数据库的预测结果存储于`./pred-dc/predictions/`目录。 - ChEMBL数据库的性能指标存储于`./train-ch/metrics/`目录。 - ChEMBL数据库的性能指标绘图存储于`./train-ch/metrics/metrics-plots/`目录。 ## 实验步骤2：训练元学习器（meta-learner） 在解压后的归档文件根目录下执行以下命令： `run-l2.bat`：基于9578个不同超参数（hyperparameters）的基础层级模型预测结果训练元学习器。模型以`./train-ch-ml/data/`目录下经处理的ChEMBL预测数据作为训练输入，对`./pred-dc-ml/data/`目录下经处理的DrugCentral预测数据进行预测。 ## 步骤2输出结果 - ChEMBL数据库的预测结果存储于`./train-ch-ml/predictions/`目录。 - DrugCentral数据库的预测结果存储于`./pred-dc-ml/predictions/`目录。 ## 补充说明 若设备搭载GPU，TensorFlow 2.10将自动调用GPU进行加速；若未搭载GPU，则回退至CPU执行。本代码已分别针对TensorFlow CPU与GPU版本完成测试。`run-all.bat`脚本可通过单次命令同时执行上述两个实验步骤。