Highlights from WG4 Hybrid Training School on MOF Computational Tools and Machine Learning

 The WG4 Hybrid Training School, organized within the EU4MOFs COST Action, was held on October 14-15, 2024, and gathered 35 on-site and 25 online researchers. The event focused on the latest advancements in data-driven and simulation techniques in the field of Metal-Organic Frameworks (MOFs).

WG4 Training school in Istanbul, Türkiye.

The meeting was held at Özyeğin University and began with a warm welcome from Prof. Dr. M. İrşadi Aksun and Assoc. Prof. Ilknur Erucar.

Prof. Dr. M. İrşadi Aksun, Vice President for Research, Innovation, and Impact (Türkiye).	Assoc. Prof. Ilknur Erucar, WG4 Leader within EU4MOFs COST Action.

The first invited speaker, Prof. Guillaume Maurin from Université de Montpellier, ICGM (France), captivated attendees with his insights into the latest developments in Machine Learning Potentials (MLPs). His presentation emphasised the new advancements in materials modelling. The workflow begins with robust data acquisition and observable descriptors. Following data acquisition, the process involves meticulous selection of ML models, alongside training and validation to ensure accuracy and reliability. Prof. Maurin detailed how rigorous model testing precedes deployment, allowing researchers to effectively utilize these models in real-world applications. The integration of ML into material science not only enhances our understanding of material behavior but also accelerates the discovery of new materials with tailored properties.

Prof. Guillaume Maurin (France)

Assoc. Prof. Sven Rogge (Belgium)

Following Prof. Guillaume Maurin’s presentation, Assoc. Prof. Sven Rogge from the Center for Molecular Modeling at Ghent University (Belgium) took the stage to delve into the world of MOFs under pressure. His talk focused on how strain engineering and compression can activate shear instability, leading to the formation of two distinct crumple zones. He presented experimental evidence demonstrating that the flexibility of soft, porous crystals can be suppressed by reducing crystallite size. Understanding the behavior of MOFs under mechanical strain opens new avenues for their application in energy storage, gas separation, and catalysis.

During a panel discussion, Prof. Seda Keskin from Koç University (Türkiye) highlighted critical issues surrounding open data in the field of MOFs. She noted the paradox of numerous publications on closed MOF models and the urgent need for open, transferable models. With more than half of existing MOF models found to be incorrect, verifying millions of database structures becomes challenging. She emphasized the importance of linking ML design with laboratory discoveries, showcasing the vital role of AI and ML in advancing material discovery.

Prof. Seda Keskin (Türkiye)

Assoc. Prof. Peyman Z. Moghadam from University College London (UK) discussed recent advancements in MOF databases and their implications for the computational discovery of MOFs for energy applications. He presented early examples of MOFs inspired by computational analysis and highlighted the role of high-throughput computational screening (HTCS) in the discovery of MOFs for gas storage and separation, showcasing the potential of computational techniques to accelerate innovation in this field.

Prof. George Froudakis (Greece)

Prof. George Froudakis from the University of Crete (Greece) explored the intersection of gas adsorption and deep learning, introducing innovative tools like a ChatGPT chemistry assistant for text mining and predicting MOF synthesis. He discussed the distinctions between machine learning and deep learning, emphasizing the role of new developments in the convolutional neural networks (CNNs) for materials modelling. He explained also concepts such as point clouds in material science, as well as descriptor-based and descriptor-free approaches.

Dr. Aydın Özcan (Türkiye)

In a subsequent panel discussion, Dr. Aydın Özcan from TÜBİTAK Marmara Research Center (Türkiye) emphasized the importance of handling missing data and applying appropriate imputation techniques. He highlighted the need for FAIR principles in MOF databases and reviewed the current state of AI models, including LLMs and generative models. He posed critical questions about whether tools like ChatGPT or Gemini truly assist researchers or primarily contribute data to larger systems. It was also highlighted to keep in mind that on-going research inherits heavy black-boxing and blind-using of current algorithms and these limitations would always be under consideration.

The next guest speaker, Dr. Jack Evans from The University of Adelaide (South Australia), discussed the advantages of dynamic materials, which offer properties that surpass static limitations. He explained how collective variables facilitate the straightforward description of phase changes and emphasized the need for enhanced sampling to explore rare or energy-intensive transitions. He also emphasized the usage of enhanced sampling methodologies to observe solid-solid phase transitions of the crystalline porous material which is a recently developing field in MOF material modelling community.

Dr. Quim Peña (Germany)

In a panel discussion, Dr. Quim Peña  from RWTH Aachen University Hospital (Germany) overviewed the state-of-the-art in nanomedicine and discussed about the challenges and perspectives of MOFs for biomedical applications. He highlighted the existing barriers that limit MOF pharmaceutical development and clinical translation, particularly in the context of drug delivery, and outlined potential niche areas where MOF properties could hold promise to generate clinical impact, such as for gas delivery or as intrinsically active metallo-therapeutic platforms.

Dr. Manuel Tsotsalas from Karlsruhe Institute of Technology (Germany) (currently in Northwestern University, USA) presented on innovative text mining techniques and their application in enhancing searches within MOF databases. He introduced the connection between his computational research and the connection of this effort with his newly established start-up company. He also emphasised how we can mitigate the entry barrier into computational and data-centric research by being a scientist from an experimental background and converting his research to more computer-based tasks.

Prof. Rochus Schmid from Ruhr-University Bochum (Germany) discussed the properties of MOFs as “soft porous crystals” with local and global structural dynamics. He emphasized that system-specific force fields (FFs) are optimal for studying MOFs, providing accuracy without bond breaking while being efficient for large length and time scales. Prof. Schmid noted that resolving the atomistic mechanisms of stimuli responses is feasible, but periodic boundary conditions can introduce artifacts. He highlighted the potential for direct “gate opening” of flexible MOF nanocrystals to study guest adsorption and transport, calling for more experimental evidence on MOF growth and surface interactions.

Student-Session

During the hands-on sessions, Nasim Samadpour, Parivash Jamshidi, and Dr. Zeynep Pinar Haşlak from Assoc. Prof. Eruçar’s MODEL research group led the training and shared their expertise. Antonios Sarikas and Michail Vlachos from Prof. George Froudakis’ group, alongside Gökhan Önder Aksu and Hasan Can Gülbalkan from Prof. Seda Keskin’s group, also contributed to the hands-on sessions. Gas adsorption simulations with rigid and flexible MOFs, ML and DL examples, drug delivery applications, polymer modelling were discussed.