fairmofsyncondition 0.1.4

A python machine learning module for for accelerating the discovery and synthesis of Metal organic frameworks

fairmofsyncondition

fairmofsyncondition is a Python module designed to predict the synthesis conditions for metal-organic frameworks (MOFs). This tool offers two main functionalities:

1. Predict Synthesis Conditions from Crystal Structures Structure: Given a crystal structure of a MOF, the model will predict the optimal set of conditions required to synthesize the specified structure.
2. Predict MOF Structures from Synthesis Conditions: If a set of reaction condition is provided, the model will predict the crystal structures of all possible MOF that can be formed under those conditions.

The model is trained on data extracted from the FAIR-MOF dataset, which is a comphrensive and carefully curated collection of MOF structures paired with their corresponding experimental synthesis conditions, building units and experimental synthetic conditions. This dataset serves as a robust foundation for accurate and reliable predictions.

Features

• Bidirectional Prediction: Whether you have a MOF structure or reaction conditions, the module can provide the corresponding synthesis conditions or possible MOF structures, respectively.
• FAIR-MOF Dataset: Utilizes a comprehensive curated dataset of MOFs with verified experimental conditions.
• User-Friendly: Easy to install and use, with minimal setup required.

Installation

The module can be installed directly from GitHub. Follow the steps below to get started:

PyPi

Simply pip install.

'''bash pip install fairmofsyncondition '''

GitHub Installation

To install fairmofsyncondition from GitHub, execute the following commands in your terminal:

# Clone the repository
git clone https://github.com/bafgreat/fairmofsyncondition.git

# Navigate into the project directory
cd fairmofsyncondition

# Install the package
pip install --upgrade pip setuptools wheel
pip install .

PYPI Installation

To install fairmofsyncondition from PYPI, simply execute the following commands in your terminal:

pip install fairmofsyncondition

Useful tool

fairmofsyncondition_syncon is a command-line tool for predicting synthetic conditions of Metal–Organic Frameworks (MOFs) directly from CIF files. It extracts organic ligands, space group information, and computes the top-5 predicted metal salts.

Quickly run command on any cif file

  fairmofsyncondition_syncon my_mof.cif

Or run and provide and outfile

  fairmofsyncondition_syncon my_mof.cif -o my_mof_report.txt

iupac2cheminfor one of the most useful tool is to directly extract cheminonformatic identifiers such as inchikey and smile strings directly from iupac names or common names. This can be achieved using iupac2cheminfor CLI as follows:

  iupac2cheminfor 'water'

or

  iupac2cheminfor -n 'water' -o filename

The out will be written by default to cheminfor.csv if no output is provided and if porvided it will be written to the name parsed.

cheminfo2iupac

Another useful tool is directly convert a smile or and inchikey their iupac name. To achieve this simply run the following commandline tool

  cheminfo2iupac -n 'O' -o filename

struct2iupac In other cases one may one to directly extract the iupac name and cheminformatic identifier of a chemical structure. The quickest way to do this is by running the following commands.

  struct2iupac XOWJUR.xyz

pg_graph from cif file or folder

One can reliably create an lmdb dataset using the following code.

  pggraph_from_cifs -i ./CIFs/ -o mof_db.lmdb

Training

To quickly train the model on the command line, simply use the train_bde CLI command. It has several helpful options to facilated training.

train_bde -h

The above command will provide all neccesarry information to train a model.

We also provide a commandline to to run optuna for searching optimal command line arguments.

find_bde_parameters -h

Machine Learning Folder

The folder machine_learning/ contains the code and Jupyter notebooks to predict the metal salt of a given MOF using Graph Neural Networks (GNNs).

• Each notebook (Ex1.ipynb, Ex2.ipynb, …, Ex10.ipynb) explores different combinations of input features such as:

  • Scherrer (grain size)
  • Microstrain (lattice distortion)
  • OMS (Open Metal Sites)
  • Atomic Number

• The notebooks share the same structure:

  1. Load the Data – import and prepare the dataset
  2. Define the GNN Model – specify the architecture
  3. Train the Model – train and save weights in tmp/ (optional)
  4. Load and Evaluate the Model – load trained weights and test performance

Note: If you only want to test the model with pre-trained weights, you can simply skip step 3 (training).

Documentation

Full documentation can be found docs.

LICENSE

This project is licensed under the MIT