Artificial Intelligence for Materials Design

In this lecture I discuss the concept of digital twin and how it can be used to optimize battery electrodes design and their manufacturing process. Different types of pioneering digital twins developed in my research group, coupling numerical simulations with AI, are presented, with illustrative application examples arising from my past and ongoing research projects, such as the ERC-funded ARTISTIC (https://www.erc-artistic.eu/). Methodologies to ensure the proper experimental calibration and validation of these digital twins are also discussed. I also present our Virtual and Mixed Reality tools designed to support the training of students, researchers and operators in the battery manufacturing field.

This talk will provide an overview of the three main approaches used within the group to aid the discovery of new materials (i) traditional bottom up construction of physical models (ii) high-throughput virtual screening (iii) machine learning (ML_. The relation between the tree approaches will be discussed including a methodology for selecting the best strategy given time and budget constraints. The challenge of dealing with experimental and highly biased datasets and the difficulty of measuring “novelty” of a given ML-based prediction will be discussed. The talk with explore how linking the prediction with the knowledge of the chemical supply chain can substantially shorten the gap between predictions and experimental verification.

Metal-organic frameworks (MOFs) are crystalline materials consisting of a metal node and an organic linker.1 By combining different metal nodes and organic linkers, chemists can synthesize infinite materials2 for applications ranging from gas separation, gas storage, sensing, catalysis, etc. This makes MOFs the ideal playground for data science. In this presentation we show how data science can support help designing MOFs for carbon capture applications. In  particular, we show how data-science methods3,4 can be used to obtain insights into questions for which conventional theory does not have an answer, such as the oxidation state of the metal5 and the heat capacity of a MOF.6 In addition, we show how data science can help us identifying the characteristics of the top-performing materials for a carbon capture process. 7
References

1    H. Furukawa, K. E. Cordova, M. O"Keeffe, and O. M. Yaghi, The Chemistry and Applications of Metal-Organic Frameworks Science 341 (6149), 974 (2013) http://dx.doi.org/10.1126/Science.1230444
2    S. Lee, B. Kim, H. Cho, H. Lee, S. Y. Lee, E. S. Cho, and J. Kim, Computational Screening of Trillions of Metal-Organic Frameworks for High-Performance Methane Storage Acs Appl Mater Inter 13 (20), 23647 (2021) http://dx.doi.org/10.1021/acsami.1c02471
3    K. M. Jablonka, D. Ongari, S. M. Moosavi, and B. Smit, Big-Data Science in Porous Materials: Materials Genomics and Machine Learning Chem. Rev. 120 (16), 8066 (2020) http://dx.doi.org/10.1021/acs.chemrev.0c00004
4    S. M. Moosavi, K. M. Jablonka, and B. Smit, The Role of Machine Learning in the Understanding and Design of Materials J. Am. Chem. Soc. 142 (48), 20273 (2020) http://dx.doi.org/10.1021/jacs.0c09105
5    K. M. Jablonka, D. Ongari, S. M. Moosavi, and B. Smit, Using collective knowledge to assign oxidation states of metal cations in metal-organic frameworks Nat. Chem. 13, 771 (2021) http://dx.doi.org/10.1038/s41557-021-00717-y
6    S. M. Moosavi, B. Á. Novotny, D. Ongari, E. Moubarak, M. Asgari, Ö. Kadioglu, C. Charalambous, A. Ortega-Guerrero, A. H. Farmahini, L. Sarkisov, S. Garcia, F. Noé, and B. Smit, A data-science approach to predict the heat capacity of nanoporous materials Nat Mater 21, 1419 (2022) http://dx.doi.org/10.1038/s41563-022-01374-3
7    C. Charalambous, E. Moubarak, J. Schilling, E. Sanchez Fernandez, J.-Y. Wang, L. Herraiz, F. Mcilwaine, K. M. Jablonka, S. M. Moosavi, J. Van Herck, G. Mouchaham, C. Serre, A. Bardow, B. Smit, and S. Garcia, Shedding Light on the Stakeholders" Perspectives for Carbon Capture. ChemRxiv  (2023) http://dx.doi.org/10.26434/chemrxiv-2023-sn90q

The progress in first-principles simulation codes and supercomputing capabilities have given birth to the so-called high-throughput (HT) ab initio approach, thus allowing for the identification of many new compounds for a variety of applications (e.g., lithium battery and photovoltaic). As a result, a number of databases have also become available online, providing access to various properties of materials, mainly ground state though. Indeed, for more complex properties (e.g., linear or higher order responses), the HT approach is still out of reach because of the required CPU time. To overcome this limitation, machine learning approaches have recently attracted much attention in the framework of materials design.
In this talk, I will review recent progress in the emerging field of materials informatics. I will briefly introduce the OPTIMADE API [1] that was developed for searching the leading materials databases (such as AFLOW, the Materials Cloud, the Materials Project, NOMAD, OQMD, ...) using the same queries. I will introduce the MODNet framework and its recent developments for predicting materials properties [2-5]. This approach, which is particularly well suited for limited datasets, relies on a feedforward neural network and the selection of physically meaningful features. Finally, I will illustrate the power of materials informatics, combining high throughput ab initio calculations and machine learning, through a few recent examples.
References
[1] C.W. Andersen et al., Sci. Data 8, 217 (2021).
[2] P. P. De Breuck, G. Hautier, and G. M. Rignanese, npj Comput. Mater. 7, 83 (2021).
[3] P. P. De Breuck, M.L. Evans, and G. M. Rignanese, J. Phys.: Condens. Matter 33, 404002 (2021).
[4] P. P. De Breuck, G. Heymans, and G. M. Rignanese, J. Mater. Inf. 2, 10 (2022).
[5] X. Liu, P. P. De Breuck, L. Wang, and G. M. Rignanese, npj Comput. Mater. 8, 233 (2022).

This talk discusses research towards the establishment of a science-driven approach to chemical machine learning (ML).[1] In many fields, ML is a fundamentally data-driven endeavor, meaning that specific databases and benchmark problems (i.e. big data) are at the center of methodological development. While this has certainly led to tremendous advances in recent years (e.g. in image generation and natural language processing), the full diversity and complexity of chemistry cannot be adequately represented by a few predefined databases. A major driver of this work is therefore the desire to build accurate, data-efficient models which do not require enormous reference datasets for training. This is because we want to be able to apply our methods to any problem of chemical interest, not just to those problems for which big data happens to be available. To this end, we explicitly incorporate chemical and physical information into the ML models[2] and integrate the data generation or selection process with the model training[3]. Several examples of this in the context of the atomistic simulation of energy materials will be discussed. 
[1] J.T. Margraf, Angew. Chemie, 62, e202219170 (2023). 
[2] K. Chen et al. Chem. Sci., 14, 4913-4922 (2023). 
[3] H. Jung et al. NPJ Comput. Mater., 9, 114 (2023).

Artificial intelligence (AI) is increasingly used by chemists and materials scientists in the design of new materials. How well do these methods work for guiding laboratory experimentation? In this talk, I will describe efforts to benchmark AI-based experiment planning methods for the exploratory synthesis of novel metal halide perovskitoid bulk solids, as well as related studies on crystallization modification additives and ligands for nanocrystals. I will describe how a combination of laboratory automation and comprehensive data capture enabled us to extract new scientific insights, develop and test new active and meta-learning algorithms, and compare a variety of algorithms in an experimental setting. Finally, I will comment on some of the limitations of existing AI methods for discovering new materials, and highlight possible research directions for improving the algorithms.

In silico materials design often involves the exploration of vast, diverse chemical spaces. While ab initio methods have been phenomenally successful in materials simulations, their scope of application has always been constrained by their high cost and poor scaling. IN this tlk, I will highlight the recent development of graph deep learning models that can universally work across the periodic table, revolutionizing our ability to explore the entire universe of materials ar unprecedented scales and accuracy. Such models can be considered as «foundational AI models» for materials chemistry, with broad applications in the dynamic simulations and discovery of materials. Finally, I will also provide some perspectives on the remaining challenges an opportunities for such models.