How to train ML model in open source code for new MOLECULES adsorption

[PhuongDongUSTH]

What would you like to report?

Hi everyone,

My name is Nguyen Phuong Dong, a Vietnamese, and a new member in our group.

Now I’m a researcher in Vietnam about CO2 reduction topic using CuS hollow nanosphere as a catalytic, and Raman laser as a trigger source of reduction reaction. My target is study the SERS Raman spectra of some of CO2 reduction products such as Methanol, Ethanol, Acetone, Propanol, Ethylene, Methane, Ethane…adsorption on CuS HKL crystal plane. To do that, the first step I have to find the best adsorption configuration of each molecular on CuS HKL plane, I plan to use this ML code in this step. After I found the best adsorption position, I will use these structures in another software to obtain the SERS Raman DFT calculation step and will archived my final expect.

From the Internet, I know our source code can do my expected. But on the online demo platform website for Catalysis just have some species for simulate the reduction mechanism, not have some molecules.

So in my post here, I would like to ask you guys support how I can train the ML model in open source code to finding the best adsorption configuration for my system {some adsorbate molecules as I mentioned above interaction with CuS HKL plane}? Could anyone help me and show me how to do that?

I very much hope that I can receive your support in my matter now.

I’m looking forward to hearing your feedback soon.

Thank you so much and have a nice day!

Best Regards,

Nguyen Phuong Dong.

[rayg1234]

Hi @PhuongDongUSTH , thanks for your interest using OCP!

The demo is not meant for any general purpose application.

For your application, I think you don't need to train any models but use pretrained OCP models to find the best adsorption configuration, you can check the tutorial here: https://fair-chem.github.io/tutorials/NRR/NRR_example.html

please also note sulfides are undersampled in OC20 training dataset so check the results carefully

[PhuongDongUSTH]

Hi @rayg1234 : Thank you so much for your comment and suggestion. I went to your suggested link. But I have one question here. How can I load my target molecules (such as Methane, HCHO, HCOOH, Methanol, Ethane, Ethylene, CH3CHO, CH3COOH, Ethanol, HOCH2CH2OH, Propanol, Acetone) in the script? I just only simply wrote the names of them as mentioned in the example of the link "H" or any suggestions for my matter here?

Because in the example link, just loading the "H" species, and this species (radical) already have in the library database of OC20, so it no problem, but for my target molecules, have not included in the library database of OC20.

[lbluque]

Hi @PhuongDongUSTH - you should be able to use your molecules loading then using ASE and creating an Adsorbate. For example if you have cif files (or any other supported format) for them:

from ase.io import read
from fairchem.data.oc.core import Adsorbate

atoms = read("HCHO.cif")
adsorbate_HCHO = Adsorbate(adsorbate_atoms=atoms)

[PhuongDongUSTH]

Hi Dr. @lbluque . Thank you so much for your comment. So is this possible to use the ".mol" or ".xyz" format of my molecules? Because those are just molecules adsorbate in my target, not crystals I cannot use the ".cif" file format.

[lbluque]

Hello @PhuongDongUSTH,

Most common structure/molecule file formats are supported by ASE (xyz and mol included). Please have a look at the ASE documentation here for further information.

Hope this helps.