(368ap) To Achieve Carbon Neutrality for Human Society, I Mainly Focus on the Electrification and Decarbonization of Separation and Chemical Process.

Figure 1: This schematic illustrates the process of predicting methane adsorption starting from crystal descriptors. Molecular building blocks and topological information are used to generate a MOF crystal. The chemical formula can be deduced from the crystal structure and used to calculate various chemical properties. These chemical properties along with the crystal descriptors are then fed into six random forest algorithms to predict six structural properties. Those structural properties and chemical properties can then be used to predict methane adsorption.

Abstract

This research developed an integrated genetic algorithm (GA) random forest (RF) machine learning algorithm (GARF) to design and screen high-performing metal-organic frameworks (MOFs) for gas adsorption. MOFs are built from repeated, coordinated units of metal clusters and organic linkers lending themselves well to being assembled in silico. Starting from merely molecular building blocks and crystal information, the algorithm can rapidly build novel MOFs and accurately predict their gas storage capabilities. MOFs can be tuned for use in specific applications including gas adsorption. By carefully selecting particular molecular building blocks, properties such as pore size can be adjusted for specific gases. The advantage of the GARF algorithm is that it can rapidly and accurately screen hundreds of thousands of MOFs in mere minutes on a personal computer. Traditional comprehensive molecular simulations can take hours to simulate adsorption for one MOF. Machine learning offers a faster way to predict adsorption, yet this technique often comes with the cost of accuracy. Previous research has shown that methane adsorption can be predicted well using structural properties and this prediction is further improved through the introduction of chemical properties [1–3]. This algorithm’s ability to rapidly identify candidate materials would be invaluable for materials discovery.

We investigated the performance of the GARF algorithm in predicting methane adsorption in MOFs. Figure 1 illustrates the process of how methane adsorption is predicted from merely building blocks and topological information. Using the crystal descriptors, a MOF structure can be generated with an accompanying chemical formula. Various chemical properties can then be calculated using the chemical formula. In order to accurately predict methane adsorption in MOFs, we required both chemical property descriptors and structural property descriptors. Since structural properties calculations require comprehensive molecular simulations, we elected to use machine learning to predict six structural properties using the crystal and chemical properties as descriptors. 50,000 hypothetical MOFs (hMOFs) from the MOFXDB database were used to train the machine learning models [4]. This database contains adsorption isotherms, structural properties, and Crystallographic Information Files (CIFs). We randomly selected 80% of the 50,000 hMOFs to train on and tested its efficacy on the remaining 20% of the data. R² values were above 0.92 for the six structural properties and most had relatively low mean absolute error percentage (MAPE) values. When trained for methane adsorption, the model had an R² value of .92 and a relatively low MAPE value of 10.2%. We could then justify the use of the RF algorithms for as the fitness evaluator for the genetic algorithm.

We then used the GARF model to evolve high-performing MOFs. The GARF evolved MOFs matched hypothetical MOFs in the top 50 highest performers in the database even when excluding the top 50 from the training dataset. We extracted the top performing evolved MOFs and were able to elucidate information about the ideal chemistries and building blocks for generating MOFs with high methane adsorption. We could then intelligently select new building blocks to add to GARF to evolve novel MOFs for methane adsorption. We plan to extend this work to discover candidate MOFs for hydrogen and carbon dioxide storage and identify the ideal chemistries for adsorption each gas. The GARF algorithm rapidly and accurately screens MOFs using minimal input information and evolves high-performing candidate MOFs for methane adsorption opening up a new avenue for materials discovery.

Research Interests

I am currently looking for a postdoctoral position with an anticipated graduation date of May 2025. My research interests include leveraging machine learning with evolutionary algorithms for improved materials discovery. While my thesis research involves discovering novel MOFs for gas adsorption, the general framework can be extended to other materials for a wide variety of applications. In a postdoctoral position, I would also like to gain experience using Density Functional Theory (DFT) to incorporate quantum properties as features to improve predictions. I can bring extensive knowledge of machine learning and evolutionary algorithms. Additionally, I have experience working in experimental research labs fabricating lithium-ion batteries and iCVD (initiated chemical vapor deposition) coated membranes for desalination applications. I would like to extend and expand the GARF (genetic algorithm / random forest) model to other applications aside from MOFs and gas adsorption.

References

1. Fernandez M, Boyd PG, Daff TD, Aghaji MZ, Woo TK. Rapid and accurate machine learning recognition of high performing metal organic frameworks for CO2 capture. J Phys Chem Lett. 2014;5:3056–60.
2. Pardakhti M, Moharreri E, Wanik D, Suib SL, Srivastava R. Machine Learning Using Combined Structural and Chemical Descriptors for Prediction of Methane Adsorption Performance of Metal Organic Frameworks (MOFs). ACS Comb Sci. 2017;19:34.
3. Beauregard N, Pardakhti M, Srivastava R. In Silico Evolution of High-Performing Metal Organic Frameworks for Methane Adsorption. J Chem Inf Model. 2021;61:3232–9.
4. Bobbitt NS, Shi K, Bucior BJ, Chen H, Tracy-Amoroso N, Li Z, et al. MOFX-DB: An Online Database of Computational Adsorption Data for Nanoporous Materials. J Chem Eng Data. 2022;68:483–98.