Toxic sulfur-containing gases such as hydrogen sulfide and sulfur oxides are serious concerns for human health, industrial safety and the environment. Developing materials that can not only detect these gases but also capture them efficiently is therefore an important goal. Yet identifying materials that balance strong adsorption with a useful sensing response remains challenging.
Now, researchers have developed a multitask deep learning framework that can simultaneously predict how strongly a material adsorbs sulfur gases and how effectively it can sense them. By combining artificial intelligence with density functional theory, or DFT, calculations, the approach could accelerate the discovery of materials for gas detection and purification.
The study, published in Artificial Intelligence & Environment, focuses on transition-metal phthalocyanines, a class of materials whose electronic properties can be tuned by changing the metal atom at their center. The researchers investigated 28 transition-metal phthalocyanine materials and four sulfur-based gases, H₂S, SO, SO₂ and SO₃, generating 78 gas-material adsorption systems.
"Gas sensing and gas adsorption are closely connected, but they are not governed by exactly the same factors," said Xiliang Yan, a corresponding author of the study. "Our goal was to build a model that can learn what these two processes have in common while also preserving the information that is unique to each task."
Traditional machine-learning approaches often predict one material property at a time. The new multitask deep learning model instead learns shared physicochemical information while using separate prediction branches for adsorption energy and sensing response. This allows the model to examine the balance between how tightly a gas binds to a material and how strongly that interaction changes the material's electrical behavior.
The approach showed clear gains in the independent test set. For classifying high and low sensing responses, the multitask model achieved an accuracy and F1 score of 0.83 and a recall of 0.88, compared with test accuracies of 0.44 to 0.63 for several conventional single-task machine-learning models. For adsorption-energy prediction, the model achieved a test-set R² of 0.86, with an RMSE of 0.46 eV and an MAE of 0.35 eV.
The calculations also identified specific material candidates. For example, Fe/Pc showed strong sensitivity toward SO, while Fe/Pc and Cr/Pc also displayed promising responses to H₂S. Other transition-metal phthalocyanines exhibited strong adsorption of SO and SO₂, highlighting their possible use as gas-removal materials.
Importantly, the AI model was designed to be interpretable rather than functioning solely as a prediction tool. Analysis of the model showed that the atomic radius and electronic properties of the transition-metal center, the electronic structure of the phthalocyanine material, and properties of the gas molecule all contribute to sensing and adsorption behavior.
"The model does more than identify promising candidates," Yan said. "It also helps reveal which physical and electronic characteristics control adsorption and sensing, providing guidance for the rational design of new materials."
The researchers note that the current model was developed from a relatively small dataset. Expanding it to additional metals, gases and adsorption configurations could improve its ability to screen previously unexplored materials.
The work demonstrates how interpretable multitask learning can connect adsorption thermodynamics with gas-sensing behavior, providing a computational strategy for designing multifunctional materials that both detect and remove hazardous gases.
===
Journal reference: Li Y; He Y; Zhou D; et al. DFT-driven multitask deep learning for sulfur gas sensing and removal by transition metal phthalocyanines. AI Environ. 2026, 1(3): 176-188. DOI: 10.66178/aie-0026-0020
https://www.the-newpress.com/aie/article/doi/10.66178/aie-0026-0020
===
About the Journal:
Artificial Intelligence & Environment is an international multidisciplinary platform for communicating advances in fundamental and applied research on the intersection of environmental science and artificial intelligence (AI). It is dedicated to serving as an innovative, efficient and professional platform for researchers in the cross-discipline fields of earth and environmental sciences, big data science and AI around the world to deliver findings from this rapidly expanding field of science. It is a peer-reviewed, open-access journal that publishes critical review, original research, rapid communication, view-point, commentary and perspective papers.
