Predicting Emission Wavelengths in Benzobisoxazole-Based OLEDs with Gradient Boosted Ensemble Models.

Area of Science:

• Organic Chemistry
• Computational Chemistry
• Materials Science

Background:

• Benzobis[1,2-d:4,5-d']oxazole (BBO) derivatives are important fluorescent emitters.
• Predicting their emission wavelengths is crucial for designing new materials.
• Existing methods may require extensive data or complex computations.

Purpose of the Study:

• To develop accurate and efficient machine learning models for predicting BBO emission wavelengths.
• To compare models trained on whole molecules versus molecular fragments.
• To provide insights into structure-property relationships using interpretable ML.

Main Methods:

• Curated a database of 50 BBO molecules with density functional theory (DFT) computed features.
• Developed two gradient-boosted ensemble machine learning (ML) models: one using whole molecules, another using fragments.
• Evaluated model performance using root-mean-square error (RMSE) and tested on unseen BBO emitters.

Main Results:

• Both ML models achieved high accuracy, with RMSEs between 30-36 nm.
• Model performance was competitive with state-of-the-art deep learning models despite using significantly less data.
• Accuracy was maintained on novel BBO emitters not included in the training set.
• Interpretable feature importance analysis revealed key structure-property relationships.

Conclusions:

• Gradient-boosted ensemble ML models provide a powerful and data-efficient approach for predicting BBO emission wavelengths.
• Fragment-based modeling shows promise for predicting properties of complex molecules.
• The study highlights the utility of interpretable ML in understanding molecular photophysics.