RMLP for organometallic catalysis

Introduction

This repository includes the structures of organometallic catalytic reaction systems discussed in the article "Accelerating the Transition State Search of Organometallic Catalysis with Reactive Machine Learning Potentials," as well as the corresponding code for transition state structure optimization using the integrated reactive machine learning potential (RMLP) model.

The contents of each folder are as follows:

1. dataset: The transition state initial guess structures of the organometallic catalytic reaction systems in the paper (including the organic ligands designed by ScaffoldCAMD, metal rhodium, and reaction substrates).
2. example: Workfolder of transition state optimization by RMLP model.
3. models: The RMLP models in the article, including MACE w/ NMS, AL MACE w/ NMS, MACE w/o NMS, PaiNN w/ NMS.
4. neuralneb: Dependency modules of PaiNN model.
5. ts_opt.py: Script for transition state optimization driven by RMLP models.

Required modules

• torch 2.5.1+cu121
• numpy 1.26.4
• xtb 22.1
• mace-torch 0.3.6
• ase 3.23.0
• sella 2.3.4
• matplotlib 3.8.2
• natsort 8.4.0
• x3dase 1.1.4

Usage tutorial

After downloading the repository using git clone or similar commands, move to the generated directory and run the following:

python ts_opt.py --model_name='mace_nms'

This command will use the MACE with NMS model to optimize the initial guess of transition state structures in ./example/input folder , and output to ./example/output.

Other arguments:

--input_path

Type: str. Specifies the XYZ format geometry path for the input.

--output_path

Type: str. Specifies the output file path.

--model_path

Type: str. Specifies the RMLP model file path.