ppODE

This repository contains the code for training a neural network model to predict drug sensitivity and synergy in triple-negative breast cancer (TNBC) through proteomics data using the ppODE (Perturbation Proteomics Ordinary Differential Equation) architecture. The model is designed to dynamically analyze perturbed proteomic time series data.

Usage

To run the script with the default parameters, use the following command:

python main.py --time_stamp_predict_drug "6_24_48" \
               --train_percent 0.7 \
               --val_percent 0.2 \
               --test_percent 0.1 \
               --taskname_prefix alltime_allpro_train_ratio_07train_02val \
               --trainval_file_prefix allcelltype_drugpair_crossdrug_ \
               --dataset_file_dir "./data/complete_data_proteo_structured_withcontrol20_0925_allproteins/" \
               --total_epoch 5000 \
               --patience 800

Arguments

• --time_stamp_predict_drug: Time stamp used to predict drug synergy. Options include "6", "24", "48", "all", or "6_24_48". Default: "6".

• --lambda_pheno: Lambda parameter for multitask learning. Default: 0.8.

• --taskname_prefix: Prefix for the task name and checkpoint files. Default: "".

• --dataset_file_dir: Directory for the dataset. Default: "./data/complete_data_proteo_structured_withcontrol20_0925_allproteins/".

• --trainval_file_prefix: Prefix for the training and validation dataset files. Default: "allcelltype_drugpair_".

• --test_file_prefix: If empty, the test set is determined by --test_percent; otherwise, the specified prefix is used for the test set. Default: "".

• --total_epoch: Total number of epochs for training. Default: 5000.

• --patience: Number of epochs to wait before early stopping. Default: 500.

• --train_percent: Percentage of data used for training. Default: 0.7.

• --val_percent: Percentage of data used for validation. Default: 0.2.

• --test_percent: Percentage of data used for testing. If this is set to 0.N, a portion of the training/validation data will be used for testing. If 0, the --test_file_prefix will define the test set. Default: 0.1.

• --cp_save_dir_best: Directory to save the best checkpoint. If empty, the best checkpoint will be selected based on training performance.

• --batch_size: Batch size for training. Default: 128.

Example

To run the ppODE model with time stamps "6_24_48", use the command:

python main.py --time_stamp_predict_drug "6_24_48" \
               --train_percent 0.7 \
               --val_percent 0.2 \
               --test_percent 0.1 \
               --taskname_prefix alltime_allpro_train_ratio_07train_02val \
               --trainval_file_prefix allcelltype_drugpair_crossdrug_ \
               --dataset_file_dir "./data/complete_data_proteo_structured_withcontrol20_0925_allproteins/" \
               --patience 800 \
               -- batch_size 64

Dependencies

Before running the script, ensure that the following Python packages are installed:

• torch
• torchdyn
• sklearn
• torcheval

You can install these dependencies using pip:

pip install torch torchdyn sklearn torcheval

License

This project is licensed under the MIT License - see the LICENSE file for details.