EasyChem-ML

EasyChem-ML is a modular structure-based machine learning tool for property, reactivity and structure prediction in (organic) chemistry, which should be easily adoptable to different problem sets.

Current Status

EasyChem-ML is developed and maintained as a PhD project by Marius Kühnemund. The whole project is neither finished nor complete. Frequent modifications can not be excluded ;) ... that's how it is in PhD.

In addition, not all modules are public available, as these were/are used for projects that have not been published yet

The framework has already been used for different research projects:

• Frederik Sandfort, Felix Strieth-Kalthoff, Marius Kühnemund, Christian Beecks, Frank Glorius, A Structure-Based Platform for Predicting Chemical Reactivity, Chem, Volume 6, Issue 6, 2020, Pages 1379-1390, ISSN 2451-9294, https://doi.org/10.1016/j.chempr.2020.02.017.

• F. Strieth-Kalthoff, F. Sandfort, M. Kühnemund, F. R. Schäfer, H. Kuchen, F.Glorius, Angew. Chem. Int. Ed. 2022, 61, e202204647; Angew. Chem. 2022, 134, e202204647. https://onlinelibrary.wiley.com/doi/full/10.1002/anie.202204647

• Pflüger, P. M., Kühnemund, M., Katzenburg, F., Kuchen, H., & Glorius, F. (2024). An evolutionary algorithm for interpretable molecular representations. Chem. https://doi.org/10.1016/j.chempr.2024.02.004

Install

For the installation, you need a working C++(For Cython) and a Rust compiler.

Linux

# Install Rust-Compiler
curl https://sh.rustup.rs -sSf | bash -s -- -y

# Install Dependencies
pip install setuptools-rust

# Install EasyChemML
git clone https://github.com/mkueh/EasyChemML.git
cd EasyChemML
pip install ./

Windows

# install Rust-Compiler
# https://www.rust-lang.org/learn/get-started

# Install Dependencies
pip install setuptools-rust

# Install EasyChemML
git clone https://github.com/mkueh/EasyChemML.git
cd EasyChemML
pip install ./

Short Example

env = EasyProjectEnvironment('TestFolder')
job_factory = Job_Factory(env)
job_runner = LocalRunner(env)

# ----------------------------------- Dataloading --------------------------------------

dataLoader = {'dreher_dataset': XLSX('Examples/DATASETS/Dreher_and_Doyle_input_data.xlsx', sheet_name='FullCV_01')}
di = DataImporter(env)
bp = di.load_data_InNewBatchPartition(dataLoader, max_chunksize=100000)

# ----------------------------------- Preprocessing --------------------------------------

molRdkit_converter = MolRdkitConverter()
molRdkit_converter.convert(bp['dreher_dataset'], columns=['Ligand', 'Additive', 'Base', 'Aryl halide'], n_jobs=10)

mff_encoder = MFF()
mff_encoder.convert(datatable=bp['dreher_dataset'], columns=['Ligand', 'Additive', 'Base', 'Aryl halide'], fp_length=16, n_jobs=64)

# ----------------------------------- Splitting --------------------------------------

split_creator = Splitcreator()
splitter_boilingpoint = ShuffleSplitter(1, 42, test_size=0.1)
splitset_boilingpoint = split_creator.generate_split(bp['dreher_dataset'], splitter_boilingpoint)

dataset_boilingpoint = Dataset(bp['dreher_dataset'],
                               name='dreher_dataset',
                               feature_col=['Ligand', 'Additive', 'Base', 'Aryl halide'],
                               target_col=['Output'],
                               split=splitset_boilingpoint, env=env)

# ----------------------------------- Metric definition-----------------------------------

r2score = R2_Score()
mae = MeanAbsoluteError()
metricStack_r = MetricStack({'r2': r2score, 'mae': mae})

catboost_r = Config(
    CatBoost_r,
    {'verbose': 50,
     'thread_count': 64,
     'allow_writing_files': False,
     'iterations': 100,
     'depth': 4}
)

# ----------------------------------- Job definition-----------------------------------

job: ModelTrainEvalJob = job_factory.create_ModelTrainEvalJob(
    'dataset_boilingpoint_0',
    dataset_boilingpoint,
    catboost_r,
    metricStack_r,
    dataset_boilingpoint.get_Splitset().
    get_outer_split(0)
)


# ----------------------------------- Training ----------------------------------------
job_runner.run_Job(job)

print(f'Test_lipo: {job.result_metric_TEST}')
print(f'Train_lipo: {job.result_metric_TRAIN}')

job.trained_Model.save_model('model_reaxys.catb')

loaded_model = CatBoost_r()
loaded_model.load_model(path='model_reaxys.catb')

X_indices = list(range(len(bp['dreher_dataset'])))
job_predict = ModelPredictJob(job_name='Predict', trained_Model=loaded_model, X=bp['dreher_dataset'],
                              X_cols=['Ligand', 'Additive', 'Base', 'Aryl halide'])
job_runner.run_Job(job_predict)

for val in job_predict.predicted_vals:
    print(str(val))


# ----------------------------------- Cleanup ----------------------------------------

env.clean()

Examples

Look inside the Example folder

Current implemented Modules

Dataloader

• CSV
• HDF5
• XLSX

Splitter

• AllTestSplitter (all Data in the Testset)
• AllTrainSplitter (all Data in the Trainset)
• RangeSplitter (define a Test/Train range)
• ShuffleSplitter (Splitts Random)

Encoder

• BertTokenizer
• Fingerprints (all Rdkit)
• MFF
• OnehotEncoder
• EasyDescriptor (support only Molmass yet)
• EVO-MFP (evolutionary trained fingerprint)

Model

• Catboost (Classification and Regression)
• Pytorch Models
  • Bert-Transformer
• RandomForest (Classification and Regression)

Jobrunner

• Localrunner (Runs the Jobs on the local machine)