Unlocking the Future of Direct Air Capture with the Open DAC 2025 Dataset

AI & Machine LearningSpatial IntelligenceAI Diagnostics

Imagine a world where direct air capture is a reality, and the Open DAC 2025 dataset is the key to unlocking it.

TL;DR
• The Open DAC 2025 dataset is a significant expansion and improvement upon ODAC23 for sorbent discovery in direct air capture.
• The dataset introduces chemical and configurational diversity through functionalized MOFs, high-energy GCMC-derived placements, and synthetically generated frameworks.
• The dataset also improves upon the accuracy of DFT calculations and the treatment of flexible MOFs in ODAC23.

Summary
The Open DAC 2025 (ODAC25) dataset expands on its predecessor by delivering nearly 70 million density functional theory (DFT) calculations across 15,000 metal‑organic frameworks (MOFs), targeting CO₂, H₂O, N₂, and O₂ adsorption in humid air. ODAC25 introduces functionalized MOFs, high‑energy grand canonical Monte Carlo (GCMC) placements, and synthetically generated structures, enhancing chemical and configurational diversity. The dataset improves DFT accuracy and flexible MOF treatment, and accompanies state‑of‑the‑art machine‑learned interatomic potentials for predicting adsorption energies and Henry’s law coefficients.
• Deploy large‑scale, high‑fidelity simulation data to train predictive models that accelerate material discovery.
• Integrate synthetic framework generation to broaden exploration beyond experimentally known structures.
• Leverage advanced interatomic potentials to reduce computational cost while maintaining accuracy.
• Use benchmark datasets to validate new algorithms before scaling to industrial pipelines.
• Share open‑source tools and models to foster collaborative progress across academia and industry.

The dataset’s breadth enables rapid screening of MOFs for direct air capture (DAC) applications, addressing the challenge of identifying effective sorbents in humid environments. By combining DFT calculations with GCMC‑derived placements, ODAC25 captures realistic adsorption scenarios, improving the relevance of predictive models. The inclusion of synthetic MOFs expands the search space, allowing exploration of novel chemistries that may outperform existing materials.

Machine‑learned potentials trained on ODAC25 demonstrate high fidelity in predicting both adsorption energies and Henry’s law coefficients, offering a computationally efficient alternative to full DFT calculations. These models can be integrated into high‑throughput screening workflows, reducing evaluation time from hours to seconds per candidate. The released potentials also provide a benchmark for future methodological advances in interatomic force field development.

ODAC25’s open‑source release, coupled with detailed documentation and reproducible workflows, establishes a community standard for DAC research. Researchers can directly compare new algorithms against a common, high‑quality dataset, accelerating innovation and reducing duplication of effort. The dataset’s comprehensive coverage of functional groups, pore sizes, and flexibility parameters ensures that models trained on it generalize well to real‑world materials.

By making the dataset and accompanying tools freely available, the project promotes inclusive access to cutting‑edge research resources. This openness encourages interdisciplinary collaboration, enabling chemists, materials scientists, and machine‑learning engineers to jointly tackle the complex problem of atmospheric CO₂ capture. The initiative exemplifies how shared knowledge and standardized benchmarks can drive rapid, scalable progress in sustainable technology development.

The Open DAC 2025 dataset has the potential to revolutionize the field of direct air capture, and we invite you to join us on this exciting journey.