Dbond

This project implements the methods described in the paper Optimizing Mirror-Image Peptide Sequence Design for Data Storage via Peptide Bond Cleavage Prediction. Its goal is to optimize mirror-image peptide sequence design through peptide bond cleavage prediction.

Project Structure

.
├── Dockerfile                # Docker environment configuration
├── LICENSE                   # License
├── PBCLA/                    # PBCLA tools and related scripts
│   ├── mgf2csv.dbond_m.py
│   ├── mgf2csv.dbond_s.py
│   ├── mgf_dataset/
│   ├── pbcla.py
│   └── utils.py
├── best_model/               # Best model weights
│   ├── dbond_m/
│   └── dbond_s/
├── checkpoint/               # Training checkpoints
│   ├── dbond_m/
│   └── dbond_s/
├── data_utils_dbond_m.py     # Data processing utilities for dbond_m
├── data_utils_dbond_s.py     # Data processing utilities for dbond_s
├── dataset/                  # Dataset files
│   ├── *.csv
├── dbond_m.py                # Main model definition for dbond_m
├── dbond_m_config/           # Configuration for dbond_m
│   └── default.yaml
├── dbond_s.py                # Main model definition for dbond_s
├── dbond_s_config/           # Configuration for dbond_s
│   ├── 1222_h_256_adam_4_b4_dn.yaml
│   └── default.yaml
├── evaluate.dbond_m.py       # Evaluation script for dbond_m
├── evaluate.dbond_s.py       # Evaluation script for dbond_s
├── multi_label_metrics.py    # Multi-label evaluation metrics
├── result/                   # Output results
│   ├── metric/
│   └── pred/
├── tensorboard/              # TensorBoard logs
│   ├── dbond_m/
│   └── dbond_s/
├── train.dbond_m.py          # Training script for dbond_m
├── train.dbond_s.py          # Training script for dbond_s
└── README.md                 # Project documentation

Main Files

• dbond_m.py / dbond_s.py: model architecture definitions
• train.dbond_m.py / train.dbond_s.py: training workflows
• evaluate.dbond_m.py / evaluate.dbond_s.py: evaluation workflows
• multi_label_metrics.py: multi-label classification metrics
• PBCLA/: peptide bond cleavage analysis tools

Quick Start

1. Environment

A prebuilt runtime image is available on Docker Hub and is the recommended way to run the project:

docker pull LoserLus/dbond_env:latest

You can also build the image locally:

docker build -t dbond_env:v0 .

2. Data Preparation

Place the dataset files in the dataset/ directory. Refer to the existing CSV files for the expected format.

The current training pipeline uses explicit train, validation, and test splits.

3. Train the Models

Example for dbond_m:

python train.dbond_m.py --config dbond_m_config/default.yaml

Example for dbond_s:

python train.dbond_s.py --config dbond_s_config/default.yaml

4. Evaluate the Models

Evaluate dbond_m

python evaluate.dbond_m.py \
	--in_model_weight_path best_model/dbond_m/2025_11_25_19_25_default_0.pt \
	--in_model_comfig_path dbond_m_config/default.yaml \
	--in_csv_to_predict_path dataset/dbond_m.test.csv \
	--out_multi_label_pred_dir result/pred/dbond_m/multi/ \
	--out_multi_label_metric_dir result/metric/dbond_m/multi/

Evaluate dbond_s

python evaluate.dbond_s.py \
	--in_model_weight_path best_model/dbond_s/2025_11_29_18_46_default_5.pt \
	--in_model_comfig_path dbond_s_config/default.yaml \
	--in_csv_to_predict_path dataset/dbond_s.test.csv \
	--in_csv_for_multi_label_path dataset/dataset.fbr.csv \
	--out_single_label_pred_dir result/pred/dbond_s/single/ \
	--out_multi_label_pred_dir result/pred/dbond_s/multi/ \
	--out_single_label_metric_dir result/metric/dbond_s/single/ \
	--out_multi_label_metric_dir result/metric/dbond_s/multi/

Argument Reference

• --in_model_weight_path: path to the model weight file
• --in_model_comfig_path: path to the model configuration file in YAML format
• --in_csv_to_predict_path: path to the input CSV file for prediction
• --in_csv_for_multi_label_path (dbond_s only): path to the CSV file used for multi-label evaluation
• --out_single_label_pred_dir (dbond_s only): output directory for single-label predictions
• --out_multi_label_pred_dir: output directory for multi-label predictions
• --out_single_label_metric_dir (dbond_s only): output directory for single-label metrics
• --out_multi_label_metric_dir: output directory for multi-label metrics

The evaluation scripts automatically save prediction results and metrics to the specified directories.

5. PBCLA Tools

The PBCLA/ directory contains peptide bond cleavage utilities for data conversion and analysis.

For more details about the data processing workflow and field definitions, see PBCLA/Readme.md.