Optymalizacja struktury cieczy jonowych pod kątem ich wykorzystania w sekwestracji ditlenku węgla
Identyfikator grantu: PT00996
Kierownik projektu: Karol Baran
Politechnika Gdańska
Wydział Chemiczny
Gdańsk
Data otwarcia: 2022-10-07
Streszczenie projektu
Number of possible combinations of atoms which can create a molecule (often referred as “chemical space”) is so big that experimental approach to evaluating substances properties is no longer efficient. There so mathematical modelling is incorporated in order to predict properties of unknown substances using knowledge of existing substances gained from experiments. This kind of modelling is known as Quantitative Structure-Property Relationship (QSPR) and highly depends on statistical, machine learning and more recently deep learning methodologies.
The aim of this project is to create artificial intelligence based tool to suggest chemical structures for potential new ionic liquids. Ionic liquids are a class of salts which are liquid at room temperature and can be used as a solvent for reactions or as sorbent materials. AI-based tool to suggest new ionic liquids composed of existing as well as novel ions will be created. There is no need for any human intervention in this process as the AI tool will work on its own. Machine learning tool to generate artificial intelligence based chemical structures that could be potential next-generation ionic liquids will be delivered as a result of the project. What is more, molecular simulation techniques will be used in order to check if substances proposed by model are valid. To ensure that molecules which are proposed by neural network are chemically feasible, molecular simulations will be performed. That way it be ensured that ILs might exist in real world.
The research plan has been divided into five tasks:
1. Comparison of different molecule representation in deep learning for ionic liquids
2. QSPR model building
3. Autoencoder model building and fine tuning for ionic liquids
4. Building reinforcement learning system
5. Verification of molecules proposed by neural network using Molecular Dynamics and ab-initio calculations
The aim of this project is to create artificial intelligence based tool to suggest chemical structures for potential new ionic liquids. Ionic liquids are a class of salts which are liquid at room temperature and can be used as a solvent for reactions or as sorbent materials. AI-based tool to suggest new ionic liquids composed of existing as well as novel ions will be created. There is no need for any human intervention in this process as the AI tool will work on its own. Machine learning tool to generate artificial intelligence based chemical structures that could be potential next-generation ionic liquids will be delivered as a result of the project. What is more, molecular simulation techniques will be used in order to check if substances proposed by model are valid. To ensure that molecules which are proposed by neural network are chemically feasible, molecular simulations will be performed. That way it be ensured that ILs might exist in real world.
The research plan has been divided into five tasks:
1. Comparison of different molecule representation in deep learning for ionic liquids
2. QSPR model building
3. Autoencoder model building and fine tuning for ionic liquids
4. Building reinforcement learning system
5. Verification of molecules proposed by neural network using Molecular Dynamics and ab-initio calculations
Publikacje
- Karol Baran, Adam Kloskowski, Graph Neural Networks and Structural Information on Ionic Liquids: A Cheminformatics Study on Molecular Physicochemical Property Prediction., JOURNAL OF PHYSICAL CHEMISTRY B / American Chemical Society 127, (2023) 10542-10555