ML-BASED PREDICTION AND GENERATION OF FLUORESCENT MOLECULES AND THEIR PROPERTIES
No Thumbnail Available
Date
2024-04-20
Authors
Nurkin, Miras
Kuatbekov, Dias
Akhmetzhanov, Kuanysh
Journal Title
Journal ISSN
Volume Title
Publisher
Nazarbayev University School of Engineering and Digital Sciences
Abstract
This report provides an in-depth overview of the “ML-based Prediction and Generation of
Fluorescence Molecules and Their Properties”, which dives into the usage of machine learning techniques
to accelerate the discovery process of new fluorescence molecules and predict their properties efficiently.
This project aims to reduce the time and costs associated with discovery and property prediction of
organic compounds.
A key component of our work was the use of generative models, including Transmol and
Chemical Language Model and the usage of deep learning architectures for property prediction. The
results of our approaches have been highly promising. We achieved nearly state-of-art performance in
predicting molecular properties and constructed a database of newly sampled molecules. It is worth
mentioning that with ensemble learning techniques, we attained an 𝑅 score of 0.956 for absorption max,
2
0.901 for emission max, and 0.73 for quantum yield. It demonstrates the high performance of our models.
During working on this project, we have learned different domains, such as RDKit for
cheminformatics, various molecular representations, and the integration of generative models with
classical machine learning algorithms to improve property prediction and synthesis.
Finally, we have developed a database containing newly sampled molecules and with their
properties, as predicted by our best performing models
Description
Keywords
Type of access: Open Access, Molecular Property Prediction, Fluorescent Molecules
Citation
Nurkin, M., Kuatbekov, D., & Akhmetzhanov, K. (2024). ML-based Prediction and Generation of Fluorescent Molecules and Their Properties. Nazarbayev University School of Engineering and Digital Sciences