Big Tech meets Small Science

© EPFL/iStock (Derick Hudson)
Meta’s high-profile push into carbon capture research generated buzz by releasing a massive open dataset of “top” materials for trapping CO₂. But now, EPFL researchers have taken a closer look, uncovering critical flaws that could steer the entire field off course.
A ripple of excitement stirred through the scientific community when Meta’s OpenDAC project released a massive dataset of quantum mechanical (DFT) calculations on materials for Direct Air Capture (DAC) applications. The scale alone was impressive—reproducing it would have cost Swiss and UK taxpayers over €25 million in computer time. Notably, the data was made openly accessible, including a list of Metal-Organic Frameworks (MOFs) that OpenDAC recommended as promising candidates for DAC. For researchers like us, this was an ideal playground—especially as OpenDAC’s work also highlighted shortcomings in the force fields we used in our PrISMa1 and USorb-DAC projects, prompting some critical self-reflection.
Our first step was to reproduce OpenDAC’s DFT results. The downloaded OpenDAC configurations gave similar binding energies for CO₂ and water, which was encouraging as we were using a different DFT program. Binding energy is a key metric in evaluating DAC materials, as it correlates with heat of adsorption and helps assess performance. However, things quickly became puzzling. OpenDAC’s results suggested that the water force field performed much better than the CO₂ one, contrary to our expectations. Even more surprising were the extremely high binding energies (exceeding 100 kJ/mol), far above the ~40–60 kJ/mol one would expect from experimental heats of adsorption.
Upon closer inspection, we traced the issue back to the input structures from the CoRe MOF database. These structures can contain chemical inconsistencies, such as missing hydrogen atoms or the removal of charged solvent molecules. These artifacts led to charged frameworks, which inflated the binding energies. In the OpenDAC project, these potential issues were noted. To our surprise, our MOFchecker tool2 identified about 40% of the structures as problematic. After removing the charged MOFs from the OpenDAC database, the binding energies of the remaining MOFs aligned much better with experimental values.
The real surprise came when we compared the binding energies predicted using our methodology with those from OpenDAC: there was essentially no correlation. This is less dramatic than it sounds—computing binding energies is sensitive to small numerical errors since it involves subtracting large energy terms. Still, our methodology consistently yielded lower energies for empty MOFs, resulting in significantly different estimates of the binding energies. While we can’t claim that ours are the true binding energies, they are better estimates than OpenDAC’s ones. These differences have major consequences. None of OpenDAC’s top-recommended MOFs met their own criteria when evaluated with our binding energies. We identified one promising material, but this one was not on OpenDAC’s list.
This endeavor, spearheaded by Xin Jin,3 highlights the power of open science; without full access to OpenDAC’s dataset, these insights would not have been possible. The lesson is clear: sharing data, even imperfect data, enables the community to learn, correct, and move forward together.
1. C. Charalambous, E. Moubarak, J. Schilling, E. Sanchez Fernandez, J.-Y. Wang, L. Herraiz, F. Mcilwaine, Shing Bo Peh, Matthew Garvin, K. M. Jablonka, S. M. Moosavi, J. Van Herck, Aysu Yurdusen Ozturk, Alireza Pourghaderi, A.-Y. Song, G. Mouchaham, C. Serre, Jeffrey A. Reimer, A. Bardow, B. Smit, and S. Garcia, A holistic platform for accelerating sorbent- based carbon capture Nature 632, 89 (2024) http://dx.doi.org/10.1038/s41586-024-07683-8
2. X. Jin, K. M. Jablonka, E. Moubarak, Y. Li, and B. Smit, MOFChecker: A Package for Validating and Correcting Metal-Organic Framework (MOF) Structures Digit Discov (2025) http://dx.doi.org/10.1039/D5DD00109A
3. X. Jin, S. Garcia, and B. Smit, Correspondence on “The Open DAC 2023 Dataset and Challenges for Sorbent Discovery in Direct Air Capture” ACS Cent. Sci. (2025) http://dx.doi.org/10.1021/acscentsci.5c00255