Repository files navigation

• ✅28th Feb 2025 - Published inNature Communications
• ✅27th Dec 2024 - Updated repository with scripts to reproduce results from the manuscript.
• 🚧TODO
  • Add prediction codes for models using 3D-structural features.
  • Add instructions to install CatPred using a Docker image.

• Google Colab Interface
• Local Installation
  • System Requirements
  • Installation
  • Prediction
  • Reproducibility
• Acknowledgements
• License
• Citations

For ease of use without any hardware requirements, a Google Colab interface is available here:tiny.cc/catpred.It contains sample data, instructions, and installation all in the Colab notebook.

If you would like to install the package on a local machine, please follow the instructions below.

• For prediction: Any machine running a Linux-based operating system is recommended.
• For training: A Linux-based operating system on a GPU-enabled machine is recommended.

Both training and prediction have been tested onUbuntu 20.04.5 LTS withNVIDIA A10 andCUDA Version: 12.0.

To train or predict with GPUs, you will need:

• CUDA >= 11.7
• cuDNN

Both options requireconda, so first install Miniconda fromhttps://conda.io/miniconda.html.

Then proceed to either option below to complete the installation. If installing the environment with conda seems to be taking too long, you can also try runningconda install -c conda-forge mamba and then replacingconda withmamba in each of the steps below.

Note for machines with GPUs: You may need to manually install a GPU-enabled version of PyTorch by following the instructionshere. If you're encountering issues with not using a GPU on your system after following the instructions below, check which version of PyTorch you have installed in your environment usingconda list | grep torch or similar. If the PyTorch line includescpu, please uninstall it usingconda remove pytorch and reinstall a GPU-enabled version using the instructions at the link above.

mkdir catpred_pipeline catpred_pipeline/resultscd catpred_pipelinewget https://catpred.s3.us-east-1.amazonaws.com/capsule_data.tar.gztar -xzf capsule_data.tar.gzgit clone https://github.com/maranasgroup/catpred.gitcd catpredconda env create -f environment.ymlconda activate catpredpip install -e.

The Jupyter Notebookbatch_demo.ipynb and the Python scriptdemo_run.py show the usage of pre-trained models for prediction.

We provide three separate ways for reproducing the results of the publication.

Estimated run time: Few minutes

Run using:

./reproduce_quick.sh

For all results pertaining to CatPred, UniKP, DLKcat, and Baseline models, this method only uses pre-trained predictions and analyses to reproduce results of the publications, including all main and supplementary figures.

Estimated run time: Up to a day depending on your GPU

Run using:

./reproduce_prediction.sh

For results pertaining to CatPred, this method uses pre-trained models to perform predictions on test sets. For results pertaining to UniKP, DLKcat, and Baseline, this method uses only pre-trained predictions and analyses to reproduce results of the publications, including all main and supplementary figures.

Estimated run time: Up to 12-14 days depending on your GPU

Run using:

./reproduce_training.sh

For all results pertaining to CatPred, UniKP, DLKcat, and Baseline models, this method trains everything from scratch. Then, it uses the trained checkpoints to make predictions and analyzes them to reproduce results of the publications, including all main and supplementary figures.

We thank the authors of the following open-source repositories:

• Chemprop - Majority of the functionality in this codebase has been inspired from theChemprop library.
• Rotary PyTorch - The rotary positional embeddings functionality for Seq-Attn. is fromRotary PyTorch.
• Progres - Protein Graph Embedding Search using pre-trained EGNN models fromProgres.

This source code is licensed under the MIT license found in theLICENSE file in the root directory of this source tree.

If you find the models useful in your research, we ask that you cite the relevant paper:

@article {Boorla2024.03.10.584340,author ={Veda Sheersh Boorla and Costas D. Maranas},title ={CatPred: A comprehensive framework for deep learning in vitro enzyme kinetic parameters kcat, Km and Ki},elocation-id ={2024.03.10.584340},year ={2024},doi ={10.1101/2024.03.10.584340},publisher ={Cold Spring Harbor Laboratory},URL ={https://www.biorxiv.org/content/early/2024/03/26/2024.03.10.584340},eprint ={https://www.biorxiv.org/content/early/2024/03/26/2024.03.10.584340.full.pdf},journal ={bioRxiv}}