Related Experiment Video
Updated: Nov 3, 2025
08:26
Synthesis of Zeolites Using the ADOR Assembly-Disassembly-Organization-Reassembly Route
Published on: April 3, 2016
13.5K
Discovering Relationships between OSDAs and Zeolites through Data Mining and Generative Neural Networks
Zach Jensen1, Soonhyoung Kwon2, Daniel Schwalbe-Koda1
1Department of Materials Science and Engineering, Massachusetts Institute of Technology, Cambridge, Massachusetts 02139, United States.
ACS Central Science
|June 3, 2021
Summary
This study uses a data-driven approach and machine learning to predict organic structure directing agents (OSDAs) for zeolite synthesis. The developed model identifies new OSDA candidates, reducing experimental effort in materials discovery.
Area of Science:
- Materials Science
- Chemistry
- Computational Science
Background:
- Organic structure directing agents (OSDAs) are essential for synthesizing micro- and mesoporous materials, particularly zeolites.
- Understanding OSDA-zeolite interactions is challenging, often relying on heuristics or costly computational methods.
- Predicting effective OSDAs for specific zeolite structures remains a significant hurdle in materials synthesis.
Purpose of the Study:
- To develop a data-driven methodology for uncovering generalized relationships between OSDAs and zeolite frameworks.
- To create a predictive model capable of suggesting novel OSDA candidates for targeted zeolite structures.
- To reduce the reliance on extensive experimentation and computationally intensive simulations in OSDA discovery.
Main Methods:
- Compiled a comprehensive database of 5,663 porous material synthesis routes from scientific literature (1966-2020).
- Employed natural language processing and text mining to extract OSDAs, zeolite phases, and gel chemistry.
- Utilized weighted holistic invariant molecular (WHIM) descriptors for OSDA structural featurization.
- Adapted a generative neural network for predicting OSDAs based on zeolite structure and gel chemistry.
- Validated generated OSDA candidates using molecular mechanics simulations.
Main Results:
- Established correlations between OSDA structural features and specific types of cage-based, small-pore zeolites.
- Successfully applied a generative neural network to identify alternative OSDA candidates for CHA and SFW zeolites.
- Demonstrated that the generated OSDA candidates are physically meaningful through molecular mechanics simulations.
- Showcased the model's ability to automate exploration of the OSDA space.
Conclusions:
- A data-driven approach, integrating NLP, featurization, and generative models, can effectively predict OSDA-zeolite relationships.
- The developed model significantly reduces the need for extensive simulation and experimentation in discovering new OSDAs.
- This approach accelerates the discovery of novel organic structure directing agents for advanced porous materials.
