RealKcat: Robust Prediction of Enzyme Variant Kinetics

Overview

Welcome to the RealKcat repository! This project provides a reproducible pipeline to predict enzyme kinetics parameters, specifically kcat and km, using curated datasets. The repository includes tools and scripts for training and inference of both kcat and km models, along with utilities for data processing, model training, and standardized prediction.

---

Quick Inference with Pretrained Model:

For a hands-on demonstration and interactive inference, use our RealKcat_Inference_Interface.ipynb notebook. Open it directly in Google Colab:

This notebook allows you to perform inference on kcat and km predictions without needing to install or configure anything locally. Simply connect to a Colab runtime, follow the provided instructions, and start exploring the RealKcat models interactively.

If you only want to make predictions with the pretrained model locally without retraining, you can use the RealKcat_Inference.ipynb notebook. This notebook offers an easy, interactive way to explore and make predictions for kcat and km values. Just provide your enzyme sequence and substrate Isomeric SMILES, and the notebook will guide you through the prediction process.

---

Retraining the Models

If you’re interested in retraining the models and reproducing the results from scratch, please follow the steps below to download and set up the required datasets.

📂 Download and Setup the Datasets

Follow these steps to download and correctly set up the datasets in the repository's data folder:

1. Download the Dataset:

   • Visit Chowdhury Lab Downloads.
   • Locate KinHub-27k (Manually-curated Enzyme Parameter Database; verified from 2158 papers) and download the dataset file (e.g., KinHub-27k.zip).

2. Move the Downloaded File:

   • Move KinHub-27k.zip to the data folder in the root directory of this repository.

3. Extract the Files into the data Directory:

   • Open a terminal or command prompt, navigate to the data directory, and unzip the dataset:

     cd path/to/RealKcat/data

     • On Linux/macOS:

       unzip KinHub-27k.zip

     • On Windows (using Command Prompt or Git Bash):

       tar -xf KinHub-27k.zip

     • Alternatively, on Windows, you can right-click KinHub-27k.zip and choose "Extract All..." to unzip directly into the data folder.

4. Verify the Extracted Files:

   • After extraction, your data folder should have the following structure:

     data/
     ├── data_split/
     ├── PafA_data/
     ├── Save_kinetic_bin_range.pkl
     ├── WT_MD_database_v1.xlsx
     ├── WT_MD_dataset.pt
     ├── WT_MD_dataset_wNeg.pt
     

5. Proceed with Training or Inference:

   • With the data set up, you can now use the provided scripts to perform training or inference.

Repository Structure

The RealKcat directory should be organized as follows:

RealKcat/
├── data/
│   ├── PafA_data/
│   │   ├── PafA_1_test_dataset_2.pt
│   │   ├── PafA_1_test_positions_2.pt
│   │   └── PafA_1_test_kcat_km_2.pt
│   ├── data_split/
│   ├── Save_kinetic_bin_range.pkl
│   ├── WT_MD_database_v1.xlsx
│   ├── WT_MD_dataset.pt
│   └── WT_MD_dataset_wNeg.pt
├── outputs/
├── scripts/
│   ├── test_ood_kcat_predict.py
│   ├── test_ood_km_predict.py
│   ├── test_PafA_kcat_predict.py
│   ├── test_PafA_km_predict.py
│   ├── train_kcat_model.py
│   └── train_km_model.py
├── src/
│   ├── data_processing.py
│   ├── evaluation.py
│   ├── model_training.py
│   └── utils.py
├── LICENSE
├── README.md
└── RealKcat_Inference.ipynb

Key Directories and Files

• data/: Contains all datasets, including training and test data, as well as supplementary files (e.g., bin range statistics for kinetic parameters).
  • PafA_data/: Datasets specifically for testing on the PafA enzyme.
  • WT_MD_*: Datasets for wild-type (WT) and mutant datasets.
• outputs/: Directory for saving model outputs, trained models, and prediction results.
• scripts/: Contains scripts for training and testing models.
  • train_kcat_model.py and train_km_model.py: Scripts for training models to predict kcat and km, respectively.
  • test_*: Scripts to run inference on test datasets for both kcat and km.
• src/: Contains utility scripts for data processing, model training, evaluation, and general utilities.
  • data_processing.py: Functions for loading and preparing datasets, standardizing data, and handling tensor data.
  • model_training.py: Functions for initializing and training XGBoost models with hyperparameters.
  • evaluation.py: Functions to evaluate model performance.
  • utils.py: Utility functions for device selection, setting seeds, etc.
• RealKcat_Inference.ipynb: Jupyter notebook for interactive inference.
• README.md: This file, providing documentation for the repository.

Installation

Prerequisites

• Python 3.8 or higher (code is compatible with Python 3.12)
• Required libraries are listed in requirements.txt.

Install Dependencies

To set up your environment, clone the repository and install the dependencies:

git clone https://github.com/TKAI-LAB-Mali/RealKcat
cd RealKcat
pip install -r requirements.txt

Usage

1. Training Models

To train models for kcat and km:

# Train kcat model
python scripts/train_kcat_model.py

# Train km model
python scripts/train_km_model.py

These scripts load datasets, standardize them, and train models with specified hyperparameters.

2. Running Inference

Run inference on test datasets using the trained models. Each test_*_predict.py script loads a model and dataset, applies standardization, and makes predictions.

Out-of-Distribution (OOD) Inference

Evaluate the model's performance on an out-of-distribution (OOD) dataset to test its robustness.

# Predict kcat on OOD dataset
python scripts/test_ood_kcat_predict.py

# Predict km on OOD dataset
python scripts/test_ood_km_predict.py

PafA Enzyme Inference

Use the PafA dataset for detailed testing on a specific enzyme.

# Predict kcat on PafA dataset
python scripts/test_PafA_kcat_predict.py

# Predict km on PafA dataset
python scripts/test_PafA_km_predict.py

3. Results and Visualization

Model outputs, trained models, and prediction results are saved as figures in the outputs/ directory.

License

📚 Citation

If you use RealKcat in your work, please cite the following:

🧬 Anna Sajeevan K, Osinuga A, B A, Ferdous S, Shahreen N, Noor MS, Koneru S, Santos-Correa LM, Salehi R, Chowdhury NB, Calderon-Lopez B, Mali A, Saha R, Chowdhury R.
Robust Prediction of Enzyme Variant Kinetics with RealKcat
bioRxiv [Preprint], 2025 Feb 15. doi: 10.1101/2025.02.10.637555
PMID: 39990461 · PMCID: PMC11844551

📄 BibTeX

@article{sajeevan2025robust,
  author = {Sajeevan, Anna K and Osinuga, Abraham and B, A and Ferdous, Sakib and Shahreen, Nabia and Noor, Mohammed Sakib and Koneru, Shashank and Santos-Correa, Laura Mariana and Salehi, Rahil and Chowdhury, Niaz Bahar and Calderon-Lopez, Brisa and Mali, Ankur and Saha, Rajib and Chowdhury, Ranjan},
  title = {Robust Prediction of Enzyme Variant Kinetics with RealKcat},
  journal = {bioRxiv},
  year = {2025},
  month = {Feb},
  day = {15},
  note = {Preprint},
  doi = {10.1101/2025.02.10.637555},
  pmid = {39990461},
  pmcid = {PMC11844551}
}

This project is licensed under the MIT License. See the `LICENSE` file for details.

## Contributing

Contributions are welcome! Please open an issue or submit a pull request for any improvements or bug fixes.