DeltaCata

DeltaCata is a siamese geometric graph neural network for direct predictions of enzyme kinetic parameter changes upon residue substitutions ($\Delta$k_cat and $\Delta$K_m). DeltaCata is easy to install and run, and is also fast and accurate (surpassing state-of-the-art kinetic parameter predictors and general-purpose protein design models).

📖 Table of Contents

• Local Installation
  • Requirements
  • Inference
  • Reproducibility
• Data Collection
• License
• Citation and Contact

💻 Local Installation

To run DeltaCata on a local machine, please follow the instructions below. DeltaCata was tested on a workstation equipped with an NVIDIA RTX 3090 GPU (CUDA 11.8) and Ubuntu 20.04.4. Environment setup typically takes about 1 hour depending on the internet bandwidth.

📦 Requirements

conda create --name DeltaCata python=3.8
conda install pytorch==2.2.0 torchvision==0.17.0 torchaudio==2.2.0 pytorch-cuda=11.8 -c pytorch -c nvidia
pip install pandas
pip install fair-esm
pip install tqdm
pip install biopython

pip install torch-scatter -f https://pytorch-geometric.com/whl/torch-2.2.0+cu118.html
pip install torch-sparse -f https://pytorch-geometric.com/whl/torch-2.2.0+cu118.html
pip install torch-cluster -f https://pytorch-geometric.com/whl/torch-2.2.0+cu118.html
pip install torch-spline-conv -f https://pytorch-geometric.com/whl/torch-2.2.0+cu118.html
pip install torch-geometric

pip install rdkit

🔍 Inference

DeltaCata supports two inference modes:

1. User-specified mutations (see input_json/example.json).
2. In silico deep mutational scanning (see input_json/example_dms.json).

Both modes require a protein sequence, a substrate SMILES string, and the corresponding PDB file path.

cd Inference/ && chmod +x ./artifacts/dssp-2.0.4/mkdssp

# Mode 1: specified mutations (example.json, typically takes 1 minute)
python inference.py --input input_json/example.json

# Mode 2: deep mutational scanning (example_dms.json, typically takes 30 minutes)
python inference.py --input input_json/example_dms.json

Required fields for input json:

• sequence: string; wild-type amino acid sequence.
• SMILES: string; SMILES of the substrate.
• pdb_path: string; path to the wild-type PDB file.
• mutant: string; in mode 1, ";" is used to separate different mutants, while "," is used to separate multiple mutated sites in a multiple-point mutant.

Note: To predict $\Delta$k_cat for a multi-substrate reaction, users can either separately predict the kinetic changes for all involved substrates and calculate their average, or select the primary target substrate as the model input while omitting common co-substrates (e.g., ATP, NADH, SAM, etc.).

🔁 Reproducibility

Using the splits in Dataset/test_dataset/ and protein structures in Dataset/test_pdbs/, the following pipeline reproduces the reported results.

cd Inference
bash reproduce.bash

📊 Data Collection

We curated data of mutation-induced changes in enzyme kinetic parameters ($\Delta$k_cat and $\Delta$K_m) from BRENDA and SABIO-RK. The pipeline for constructing the DeltaCata-DB dataset is provided in Data_collection/.

For quick use, ready-to-use datasets are available in Dataset/ as delta_kcat.csv and delta_km.csv.

📄 License

The source code of DeltaCata is licensed under the Attribution-NonCommercial-NoDerivatives 4.0 International license given in the LICENSE file.

🗞️ Citation and Contact

Citation: Coming soon.

Contact:
Qianmu Yuan (yuanqm3@mail3.sysu.edu.cn)
Yuedong Yang (yangyd25@mail.sysu.edu.cn)