Deep Learning-Driven Molecular Generation and Electrochemical Property Prediction for Optimal Electrolyte Additive Design
Recently, generative models have rapidly advanced and are being applied to various domains beyond vision and large language models (LLMs). In the field of chemistry and molecular generation, deep learning-based models are increasingly utilized to reduce experimental exploration and research costs. I...
Saved in:
| Main Authors: | , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
MDPI AG
2025-03-01
|
| Series: | Applied Sciences |
| Subjects: | |
| Online Access: | https://www.mdpi.com/2076-3417/15/7/3640 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Recently, generative models have rapidly advanced and are being applied to various domains beyond vision and large language models (LLMs). In the field of chemistry and molecular generation, deep learning-based models are increasingly utilized to reduce experimental exploration and research costs. In this study, we conducted research on Variational Autoencoder-based molecular generation and property prediction to screen for optimal molecules in the design of electrolyte additives for lithium-ion batteries. Using a dataset composed of promising electrolyte additive candidate molecules, we generated new molecules and predicted HOMO and LUMO values, which are key factors in electrolyte additive design. For approximately 1000 newly generated electrolyte additive candidate molecules, we performed DFT calculations to obtain HOMO and LUMO values and calculated the mean absolute error (MAE) between the predicted values from the trained model and the DFT-calculated values. As a result, the model demonstrated exceptionally low errors of approximately 0.04996 eV (HOMO) and 0.06895 eV (LUMO), respectively. This means that battery experts can receive recommendations for new molecules, refer to their predicted HOMO and LUMO values, and select potential electrolyte additives for further validation through experiments. By replacing the traditional electrolyte additive development process with deep learning models, this method has the potential to significantly reduce the overall development time and improve efficiency. |
|---|---|
| ISSN: | 2076-3417 |