Of all greenhouse gases, carbon dioxide is the greatest contributor to global warming. According to the Intergovernmental Panel on Climate Change, without action, the average global temperature will rise by about 1.5°C by 2100. Finding effective ways to capture and store carbon dioxide has become a challenge for researchers and industry focused on combating global warming. Amir Bharati Farimani(opens in new window) I have been working to change that.
“Machine learning models have the potential to discover new compounds and materials to combat global warming,” explains Bharati Farimani, an assistant professor at the university. mechanical engineering(opens in new window) at Carnegie Mellon University. “Machine learning models can achieve accurate and efficient virtual screening of CO.”2 It can also be a storage candidate and generate favorable compounds that did not exist before. ”
Barati Farimani made breakthrough progress using machine learning to identify ionic liquid molecules. Ionic liquids (ILs) are a class of molten salts that remain liquid at room temperature and have high chemical stability and high CO2.2 Highly soluble, making it an ideal candidate for CO2 Storage. The properties of IL are largely determined by the combination of ions. However, such combinatorial possibilities of cations and anions make it very difficult to exhaust the IL design space to achieve efficient CO.2 Storage from traditional experiments.
Machine learning is often used in drug discovery to create so-called molecular fingerprints alongside graph neural networks (GNNs), which treat molecules as graphs and use matrices to identify molecular bonds and related properties. . Barati Farimani was the first to develop both his fingerprint-based ML model and GNNs that can predict CO.2 Absorption in ionic liquids.
“Our GNN method achieves excellent accuracy in CO prediction.2 solubility in ionic liquids,” Bharati Farimani said. “Unlike his previous ML methods, which rely on hand-crafted features, GNNs learn features directly from the molecular graph.”
Understanding how machine learning models make decisions is as important as the molecular properties they identify. This description provides researchers with further insight into how the structure of molecules affects the properties of ionic liquids from a data-driven perspective. For example, Barati Farmimani’s team discovered a molecular fragment that physically interacts with CO.2 Not as important as those with chemical interactions. In addition, those with low amounts of hydrogen bound to nitrogen may be advantageous in forming stable chemical interactions with CO.2.
These findings will enable researchers to advise on the design of new and efficient ionic liquids for CO.2 future storage.
