To help you get started, we've collected the most common ways that chemprop is being used within popular public projects.

wengong-jin / chemprop / chemprop / train / train.py View on Github

num_iters = len(data) if args.last_batch else len(data) // args.batch_size * args.batch_size

    if args.parallel_featurization:
        batch_queue = Queue(args.batch_queue_max_size)
        exit_queue = Queue(1)
        batch_process = Process(target=async_mol2graph, args=(batch_queue, data, args, num_iters, args.batch_size, exit_queue, args.last_batch))
        batch_process.start()
        currently_loaded_batches = []

    iter_size = 1 if args.maml else args.batch_size

    for i in trange(0, num_iters, iter_size):
        if args.moe:
            if not args.batch_domain_encs:
                model.compute_domain_encs(train_smiles)  # want to recompute every batch
            mol_batch = [MoleculeDataset(d[i:i + args.batch_size]) for d in data]
            train_batch, train_targets = [], []
            for b in mol_batch:
                tb, tt = b.smiles(), b.targets()
                train_batch.append(tb)
                train_targets.append(tt)
            test_batch = test_smiles[i:i + args.batch_size]
            loss = model.compute_loss(train_batch, train_targets, test_batch)
            model.zero_grad()

            loss_sum += loss.item()
            iter_count += len(mol_batch)
        elif args.maml:
            task_train_data, task_test_data, task_idx = data.sample_maml_task(args)
            mol_batch = task_test_data
            smiles_batch, features_batch, target_batch = task_train_data.smiles(), task_train_data.features(), task_train_data.targets(task_idx)
            # no mask since we only picked data points that have the desired target

View more ways to use chemprop

Chemprop

Chemprop is a repository containing message passing neural networks for molecular property prediction.

Documentation can be found here.

There are tutorial notebooks in the examples/ directory.

Chemprop recently underwent a ground-up rewrite and new major release (v2.0.0). A helpful transition guide from Chemprop v1 to v2 can be found here. This includes a side-by-side comparison of CLI argument options, a list of which arguments will be implemented in later versions of v2, and a list of changes to default hyperparameters.

License: Chemprop is free to use under the MIT License. The Chemprop logo is free to use under CC0 1.0.

References: Please cite the appropriate papers if Chemprop is helpful to your research.

• Chemprop was initially described in the papers Analyzing Learned Molecular Representations for Property Prediction for molecules and Machine Learning of Reaction Properties via Learned Representations of the Condensed Graph of Reaction for reactions.
• The interpretation functionality (available in v1, but not yet implemented in v2) is based on the paper Multi-Objective Molecule Generation using Interpretable Substructures.
• Chemprop now has its own dedicated manuscript that describes and benchmarks it in more detail: Chemprop: A Machine Learning Package for Chemical Property Prediction.
• A paper describing and benchmarking the changes in v2.0.0 is forthcoming.

Selected Applications: Chemprop has been successfully used in the following works.

• A Deep Learning Approach to Antibiotic Discovery - Cell (2020): Chemprop was used to predict antibiotic activity against E. coli, leading to the discovery of Halicin, a novel antibiotic candidate. Model checkpoints are availabile on Zenodo.
• Discovery of a structural class of antibiotics with explainable deep learning - Nature (2023): Identified a structural class of antibiotics selective against methicillin-resistant S. aureus (MRSA) and vancomycin-resistant enterococci using ensembles of Chemprop models, and explained results using Chemprop's interpret method.
• ADMET-AI: A machine learning ADMET platform for evaluation of large-scale chemical libraries: Chemprop was trained on 41 absorption, distribution, metabolism, excretion, and toxicity (ADMET) datasets from the Therapeutics Data Commons. The Chemprop models in ADMET-AI are available both as a web server at admet.ai.greenstonebio.com and as a Python package at github.com/swansonk14/admet_ai.
• A more extensive list of successful Chemprop applications is given in our 2023 paper

Version 1.x

For users who have not yet made the switch to Chemprop v2.0, please reference the following resources.

v1 Documentation

• Documentation of Chemprop v1 is available here. Note that the content of this site is several versions behind the final v1 release (v1.7.1) and does not cover the full scope of features available in chemprop v1.
• The v1 README is the best source for documentation on more recently-added features.
• Please also see descriptions of all the possible command line arguments in the v1 args.py file.

v1 Tutorials and Examples

• Benchmark scripts - scripts from our 2023 paper, providing examples of many features using Chemprop v1.6.1
• ACS Fall 2023 Workshop - presentation, interactive demo, exercises on Google Colab with solution key
• Google Colab notebook - several examples, intended to be run in Google Colab rather than as a Jupyter notebook on your local machine
• nanoHUB tool - a notebook of examples similar to the Colab notebook above, doesn't require any installation
  • YouTube video - lecture accompanying nanoHUB tool
• These slides provide a Chemprop tutorial and highlight additions as of April 28th, 2020

v1 Known Issues

We have discontinued support for v1 since v2 has been released, but we still appreciate v1 bug reports and will tag them as v1-wontfix so the community can find them easily.