Polaris is a benchmarking platform for the machine-learning drug discovery community. It helps researchers share, find, and use datasets and benchmarks, with the goal of improving how computational methods are evaluated and supporting their impact on real-world drug discovery. The platform offers a central hub for accessing curated scientific data, loading datasets and benchmarks with its Python package, evaluating model predictions, and sharing results.

Its resources focus on small molecules and include data relevant to phenomics, chemical binding, and ADME properties. Featured examples include RxRx3-core, a compact collection of Cell Painting images, embeddings, and perturbation labels; BELKA-v1, a large dataset of small molecules screened against three protein targets; and adme-fang-v1, covering six in vitro absorption, distribution, metabolism, and excretion endpoints. Polaris also promotes guidance for dataset curation and method evaluation, alongside recommended benchmarks developed through cross-industry collaboration among pharmaceutical, biotechnology, and research organizations.

Users can explore available resources, contribute to the community’s benchmarking efforts, and sign up for platform updates