🧪 Reaction Prediction Model

This project provides a reaction prediction model based on a deep learning framework using graph neural networks (GNN). It predicts whether a given pair of reactants is likely to undergo a reaction. The model accepts input as pairs of SMILES strings and outputs the predicted reaction probability.

---

📂 Project Structure

reaction-predictor/
├── predict.py                # Command-line interface for prediction
├── train.py                  # Training script (optional, not required for inference)
├── RxnPred/                  # Core model and data utilities
│   ├── model.py
│   ├── getTfRecord.py
│   └── ...
├── Model/
│   └── graph\_structure\_reaction/
│       ├── *.json            # Model config files
│       ├── model\_*.ckpt      # Trained model weights
│       └── isotonic\_model.joblib  # Isotonic regression calibration model
├── demo.csv                  # Example input file
├── demo\_out.csv              # Example output file
└── README.md                 # Documentation

---

🚀 Installation

1. Clone the repository

git clone https://github.com/ZhuMetLab/ReactionPredictor.git
cd ReactionPredictor

2. Install dependencies

We recommend using a conda environment:

conda create -n RxnPred python=3.10
conda activate RxnPred

pip install tensorflow==2.12.0 pandas tqdm joblib rdkit==2022.9.1 bayesian-optimization molmass

---

📥 Input Format

Input should be a CSV file with two columns:

• SMILES1: SMILES string of the first metabolite
• SMILES2: SMILES string of the second metabolite

Example (demo.csv):

SMILES1,SMILES2
C(CC(=O)O)[C@@H](C(=O)O)N,CC(=O)N[C@@H](CCC(=O)O)C(=O)O
CCCCC,C(C(=O)O)N
CC(=O)SCCNC(=O)CCNC(=O)[C@@H](C(C)(C)COP(=O)(O)OP(=O)(O)OC[C@@H]1[C@H]([C@H]([C@@H](O1)N2C=NC3=C(N=CN=C32)N)O)OP(=O)(O)O)O,CC(C)(COP(=O)(O)OP(=O)(O)OC[C@@H]1[C@H]([C@H]([C@@H](O1)N2C=NC3=C(N=CN=C32)N)O)OP(=O)(O)O)[C@H](C(=O)NCCC(=O)NCCS)O
ABCDE,C(C(=O)O)N
C(CS(=O)(=O)O)N,C(CS(=O)(=O)O)NC(=O)CO
C(CCNCCCN)CN,C(CCNCCCN)CNCCCN
C(CCNCCCN)CNCCCN,CCCCCC
C([C@@H]1[C@H]([C@@H]([C@H](C(O1)O)O)O)O)O,C([C@H]([C@H]([C@@H]([C@H](C=O)O)O)O)O)OP(=O)(O)O
COCOCOCOCOCOCOCOCOCOCOC,C[C@H](CCCC(C)C)[C@H]1CC[C@@H]2[C@@]1(CC[C@H]3[C@H]2CC=C4[C@@]3(CC[C@@H](C4)O)C)C
C(C(=O)O)N,TiO2

Note: Invalid or unparseable SMILES will be flagged as ERROR in the output.

---

📤 Output Format

The output CSV will contain the same SMILES pairs with an additional score column indicating predicted reaction probability (0–1), or "ERROR" for invalid inputs.

---

🧠 Running Prediction

You can run prediction from the command line:

python predict.py --input demo.csv --output demo_out.csv

Optional arguments:

• --batch_size: Batch size for prediction (default: 64)
• --keep_tfrecord: If specified, temporary TFRecord files will be preserved

Example with options:

python predict.py --input demo.csv --output demo_out.csv --batch_size 128 --keep_tfrecord

---

🧪 Model Info

• Architecture: GNN-based model using graph convolutions and dense layers
• Input: Metabolite SMILES pairs
• Output: Reaction probability

---

⚠️ Notes

• Input SMILES must be valid; otherwise the line will be marked as "ERROR".
• Temporary files (.tfrecord) are deleted by default unless --keep_tfrecord is specified.
• This repo currently supports prediction only. For training or model fine-tuning, refer to train.py.

---

📬 Contact

For questions, feel free to open an issue or contact the authors at [zhanghs@sioc.ac.cn].

---

⚠️ License

This work is licensed under the Attribution-NonCommercial-NoDerivatives 4.0 International (CC BY-NC-ND 4.0). See LICENSE for details.