Working Group 4

Predicting the structure and properties of MOF is challenging due to the abundance of reported MOF structures, the lack of standard databases that enable trend analysis at each specific length scale (nano, meso, and macro), at the production level (mg, g, kg, and tonnes), and for applications such as cancer nanomedicine, energy storage, and water purification. As a result, the optimal design and construction of highly effective MOF structures offers enormous technological value. This issue is dependent on the essential characteristics that are unique to each application of interest (i.e., closely tied to the inherent constraints of each dimension: nano, meso, and macro). Within EU4MOFs, we aim to strengthen such relationships to improve the control and prediction of properties such as i) long-term storage stability, reproducibility, efficacy, and toxicity of MOF nanoparticle formulations for nanomedicine, ii) anisotropic orientation, conductivity, efficiency, and material stability for energy storage, and iii) reusability, robustness and low cost of macroscopic materials for water treatment. In addition, we want to understand how these properties could potentially be affected in the large-scale production of nano-, meso- and macroscopic MOF materials. For this aim, we want to first identify the key functional- and technology-promoting properties, specific for each of the three applications by utilizing high throughput computational screening methods and machine learning tools.

EU4MOFs will consolidate high-throughput computational screening for MOFs in (cancer) nanomedicine, energy storage, and wastewater treatment by also integrating machine (and deep) learning tools that maximize structure-performance relationships. By adequately employing artificial intelligence, optimized physical and functional properties (e.g., size, shape, charge, surface coating, stability, conductivity, toxicity) could be predicted and customized at all length scales. Although recent studies on machine learning approaches for gas storage and separation using MOFs are very promising, the results in fields such as (cancer) nanomedicine or wastewater remain suboptimal due to insufficient data resulting from a lack of standard MOF databases. The progress in computational methods and technologies have led to algorithms and predictive models with high accuracy (i.e., even beyond 90-95%). However, the remaining 5-10% margin of error (partly due to insufficient data) can result in millions of (unacceptable) potential wrong decisions for e.g., health-related applications. Thus, taking advantage of the involvement of multiple groups in the Action, EU4MOFs aims to generate and establish guidelines for high-fidelity, robust, and comparable (meta)data to systematically train current computational systems and enable more accurate structure property- function predictions. This task goes beyond the state-of-the-art because will be focused on

i) structuring data and establishing MOF structure-performance trend analysis,
ii) assessing validity of predictions for certain MOF properties within the three main applications, and
iii) evaluating potential technological value by predicting factors like space-time yield (STY).

Objective and Tasks:

• Establishing a standardized database for MOF materials and their functional properties (performance) in the three main applications for machine learning. The key to foster integrated computationally assisted high-throughput screening and highly accurate machine learning tool-based predictive algorithms to optimize MOF structure design relies on the generation of accurate (meta)data and the collection of MOF-based database. Standardized data (from WG3) on physical characteristics, such as porosity, volume, structure, but also performance (e.g., toxicity, conductivity, porosity, and chemical stability) under different conditions will be collected. Additionally, establishing consensus and guidelines for the generation of high-quality (meta)data will be a priority.
• Assessing the predictive quality of machine learning tools and algorithms by integrating existing and novel MOF data. To do that, the working group will i) systematically analyze and collect already published (and potentially available unpublished due to unsuccessful outcomes) MOF data for the three applications, and ii) integrate novel data from WG1-3 to train models and strengthen the quality of the algorithms and the accuracy of the prediction.

Milestones (M) and Deliverables (D)
M4.1: Computational methods for optimized design and structural screening of MOFs established. (Month 24).
M4.2: Validated machine learning models and algorithms for MOF structure-property (e.g., toxicity, stability, conductivity, absorptivity) prediction. (Month 48)
D4.1: Training school/workshop on focused on high-throughput screening, modelling, and artificial intelligence (machine learning) tools for materials science. (Month 12)
D4.2: Perspective on machine learning integrated to functional porous (nano)materials (Month 36).
D4.3: Expandable online database connecting measured material properties with synthesis methods and modelling data. (Month 48)

WG4 Leader: Dr. İlknur ERUCAR FINDIKCI (ilknur.erucar@ozyegin.edu.tr)

WG4 Co-Leader: Prof. George FROUDAKIS (frudakis@uoc.gr)