plinder, short for protein ligand interactions dataset and evaluation resource, is a dataset and resource for training and evaluation of protein-ligand docking algorithms.

It is a comprehensive, annotated, high quality dataset, including:

• > 400k PLI systems across > 11k SCOP domains and > 50k unique small molecules
• 500+ annotations for each system, including protein and ligand properties, quality, matched molecular series and more
• Automated curation pipeline to keep up with the PDB
• 14 PLI metrics and over 20 billion similarity scores
• Unbound (apo) and predicted Alphafold2 structures linked to holo systems
• train-val-test splits and ability to tune splitting based on the learning task
• Robust evaluation harness to simplify and standard performance comparison between models

The plinder project is a community effort, launched by the University of Basel, SIB Swiss Institute of Bioinformatics, VantAI, NVIDIA, MIT CSAIL, and will be regularly updated We highly welcome contributions! If you find plinder useful, please see the citation file for details on how to cite.

To accelerate community adoption, PLINDER will be used as the field’s new Protein-Ligand interaction dataset standard as part of an exciting competition at the upcoming 2024 Machine Learning in Structural Biology (MLSB) Workshop at NeurIPS, one of the fields’ premiere academic gatherings, which will be announced shortly.

Please use a virtual environment for the plinder project. We recommend the miniforge environment manager.

NOTE: We currently only support a Linux environment. plinder uses openstructure for some of its functionality and is available from the aivant conda channel using conda install aivant::openstructure, but it is only built targeting Linux architectures. For MacOS users, please see the relevant docker resources below.

The plinder package can be obtained from GitHub:

git clone https://github.com/plinder-org/plinder.git
cd plinder
mamba env create -f environment.yml
mamba activate plinder
pip install .

Or with a development installation:

cd plinder
pip install -e '.[dev]'

Using the plinder.core API, you can transparently and lazily download and interact with most of the components of the dataset. However, if you prefer to use the dataset directly, you can fetch it using gsutil. For portions of the code that still assume an already available pre-existing dataset, you will need to use gsutil.

To download the manifest of available versions:

gsutil -m cp -r gs://plinder/manifest.md .

Then you can choose to download a particular README version or download the entire dataset with:

export PLINDER_RELEASE=2024-06
export PLINDER_ITERATION=v2
gsutil -m cp -r gs://plinder/${PLINDER_RELEASE}/${PLINDER_ITERATION}/* ~/.local/share/plinder/${PLINDER_RELEASE}/${PLINDER_ITERATION}/

View the hierarchy of a particular version of the dataset with:

gsutil ls gs://plinder/2024-06/v2/

Note that gsutil also supports glob patterns like:

gsutil ls 'gs://plinder/2024-04/v1/fingerprints/*'

This will list the files at the given path in cloud storage.

NOTE: We keep the default PLINDER_RELEASE and PLINDER_ITERATION in the source code up-to-date with the latest version of the dataset, so if you plan to use a different version, be sure to set the environment variables accordingly.

We version the plinder dataset with two controls:

• PLINDER_RELEASE: the month stamp of the last RCSB sync
• PLINDER_ITERATION: value that enables iterative development within a release

We version the plinder application using an automated semantic versioning scheme based on the git commit history. The plinder.data package is responsible for generating a dataset release and the plinder.core package makes it easy to interact with the dataset.

As part of plinder resource we also provide train, validation and test splits that are curated to minimize the information leakage based on protein-ligand interaction similarity. In addition, we have prioritized the systems that has a linked experimental apo structure or matched molecular series to support realistic inference scenarios for hit discovery and optimization. Finally, a particular care is taken for test set that is further prioritized to contain high quality structures to provide unambiguous ground-truths for performance benchmarking.

Moreover, as we enticipate this resource to be used for benchmarking a wide range of methods, including those simultaneously predicting protein structure (aka. co-folding) or those generating novel ligand structures, we further stratified test (by novel ligand, pocket, protein or all) to cover a wide range of tasks.

Our latest test split contains:

See the plinder.eval docs for more details.

Dataloader is currently under construction.

See the plinder.data docs for more details.

We are currently working on the following:

• Implementing the Dataloader
• Establishing a leaderboard
• Improving the documentation and examples

This code is split into 4 sub-packages

• plinder.core: core data structures for interacting with and loading the dataset.
• plinder.data: core code for generating the dataset
• plinder.eval: evaluation harness for the dataset that takes as an input predicted and ground truth structures in a pre-determined folder structure and returns a leaderboard-ready set of entries
• plinder.methods: implementations of the methods in the leaderboard that leverage plinder-primitives for training & running

See the End-to-end pipeline description for technical details about the dataset generation.

Package documentation, including API documentation, example notebooks, and supplementary guides, are made available.

To develop and test changes to the source code, please use a development installation:

git clone https://github.com/plinder-org/plinder.git
# OR
git clone git@github.com:plinder-org/plinder.git
cd plinder
mamba env create -f environment.yml
mamba activate plinder
pip install -e '.[dev]'

Please install pre-commit hooks (the same checks will run in CI):

pre-commit install

Test linting checks:

tox -e lint

Run typing checks:

tox -e type

Run the test suite:

tox -e test

We lint with ruff. See tox.ini and .pre-commit-config.yaml for details.

In order to change log levels in plinder, please set:

export PLINDER_LOG_LEVEL=10

This is a community effort and as such we highly encourage contributions.

We publish the plinder project as a docker container, to ensure the highest level of compatibility with non-Linux platforms. See the relevant docker resources here for more details:

• docker-compose.yml: defines a base image, the plinder "app" and a test container
• dockerfiles/base/: contains the files for the base image
• dockerfiles/main/: contains the files for the plinder "app" image

The CI workflow will automatically semver bump the plinder version and publish the plinder image to the GitHub Container Registry on merges to main. Control over the version bumping semantics is handled by inspecting the commit history since the previous release:

• If bumpversion skip is present in the commit message, the version will not be bumped
• If bumpversion major is present in the commit message, the major version will be bumped
• If bumpversion minor is present in the commit message, the minor version will be bumped
• If bumpversion patch is present in the commit message (or nothing is found), the patch version will be bumped

NOTE: The CI workflow will use the most recent match in the commit history to make its decision.

Durairaj, Janani, Yusuf Adeshina, Zhonglin Cao, Xuejin Zhang, Vladas Oleinikovas, Thomas Duignan, Zachary McClure, Xavier Robin, Emanuele Rossi, Guoqing Zhou, Srimukh Prasad Veccham, Clemens Isert, Yuxing Peng, Prabindh Sundareson, Mehmet Akdel, Gabriele Corso, Hannes Stärk, Zachary Wayne Carpenter, Michael M. Bronstein, Emine Kucukbenli, Torsten Schwede, Luca Naef. 2024. “PLINDER: The Protein-Ligand Interactions Dataset and Evaluation Resource.” bioRxiv ICML'24 ML4LMS