Predicting Fluorescence Emission Wavelengths and Quantum Yields via Machine Learning.

Area of Science:

• Computational chemistry
• Materials science
• Machine learning

Background:

• Predicting photophysical properties of fluorescent organic materials is crucial but computationally and experimentally intensive.
• Screening large numbers of potential fluorophore molecules in various solvents is a significant challenge.

Purpose of the Study:

• Develop machine learning (ML) algorithms for rapid and accurate prediction of molecular fluorescence properties.
• Focus on predicting emission wavelengths (WLs) and quantum yields (QYs) for organic fluorophores.

Main Methods:

• Utilized the Deep4Chem database containing 20,236 chromophore-solvent combinations.
• Employed chemical descriptors and SMILES fingerprints as input features for ML models.
• Applied the Shapley additive explanations (SHAP) technique for result interpretation.

Main Results:

• Random Forest model achieved a root-mean-square error (RMSE) of 28.8 nm for WLs and 0.19 for QYs on the test set.
• SHAP analysis confirmed the importance of chromophore-related properties in predictions.
• Generated two new databases by predicting missing WL and QY data.

Conclusions:

• Developed effective ML models for predicting fluorescence emission wavelengths and quantum yields.
• The ML models successfully predicted missing data, creating valuable resources for materials discovery.
• Validated the models' performance on molecules outside the original training dataset.