Oct 27 – 30, 2024
Achat Hotel Karlsruhe City
Europe/Berlin timezone

MOFGalaxyNet: Bridging AI Modeling and Social Networking for ‎Predicting Guest Accessibility in Metal-Organic Frameworks

Oct 30, 2024, 9:45 AM
20m
Kurfürstensaal (Achat Hotel)

Kurfürstensaal

Achat Hotel

Talk Machine learning (ML) applications using existing data repositories Session - click on "Detailed view" on the top right to see all contributions

Speaker

Mehrdad Jalali (Karlsruhe Institute of Technology)

Description

Integrating artificial intelligence (AI) with metal-organic frameworks (MOFs) and highly versatile and structurally ‎diverse materials heralds a new era in material science, offering groundbreaking solutions to longstanding ‎challenges in engineering and data analytics. MOFs, known for their exceptional porosity and customizable ‎frameworks, have shown promising applications across various fields, including gas storage, separation ‎processes, catalysis, and drug delivery. However, accurately predicting and analyzing the accessibility of guest ‎molecules within these frameworks is a critical determinant of their functional performance.‎

Our research introduces MOFGalaxyNet1, a novel AI framework designed to significantly enhance the prediction ‎and analysis of guest molecule accessibility within MOFs, moving beyond the limitations of traditional ‎methodologies. By employing a unique combination of social network analysis (SNA) and graph convolutional ‎networks (GCNs), MOFGalaxyNet offers a new perspective on the structural analysis of MOFs. This approach ‎treats MOFs as dynamic networks of nodes and edges, where nodes represent the organic linkers or metal ions, ‎and edges depict the connections between them. This paradigm allows for an in-depth understanding of the ‎structural features governing guest molecule accessibility. It facilitates the transformation of MOFs' structural ‎data into a numerical vector representation encapsulating their connectivity and topology.‎

One cornerstone innovation of MOFGalaxyNet is the application of GCNs to predict the pore-limiting diameter ‎‎(PLD) of MOFs, a critical parameter for determining guest accessibility. This capability significantly advances ‎existing machine learning models, offering a faster, more efficient means of screening MOFs for specific ‎applications. Through a comprehensive dataset of various MOFs characterized by their unique structural ‎properties, MOFGalaxyNet has demonstrated superior performance in predicting PLDs with remarkable ‎accuracy.‎

The implications of our findings extend beyond the immediate advancements in MOF research. By highlighting ‎the potential of AI to revolutionize material discovery and characterization, MOFGalaxyNet exemplifies the ‎broader significance of integrating AI with material science.

Primary author

Mehrdad Jalali (Karlsruhe Institute of Technology)

Co-author

Prof. Christof Wöll (Karlsruhe Institute of Technology)

Presentation materials

There are no materials yet.