Exploring Optimized Organic Fluorophore Search through Experimental Data-Driven Adaptive β‑VAE.

Area of Science:

• Organic Materials Science
• Computational Chemistry
• Photophysics

Background:

• Designing organic fluorescent molecules with specific optical properties is challenging.
• Existing inverse design methods often focus on theoretical properties, not experimental ones.

Purpose of the Study:

• To introduce a novel inverse design strategy for optimizing specific experimental properties of organic fluorescent molecules.
• To enhance the robustness and diversity of generated molecules.

Main Methods:

• Employed an adaptive beta-variational autoencoder (adaptive β-VAE).
• Combined the adaptive β-VAE with a latent vector-based prediction model.
• Dynamically tuned the Kullback-Leibler divergence scaling factor (β) and used a separate training strategy.

Main Results:

• Latent vectors from the adaptive β-VAE effectively predict experimental properties like fluorescence energy and quantum yield.
• The method enhances generator robustness and molecular diversity.
• A gradient-guided search framework in latent space identified optimal molecular designs.

Conclusions:

• The developed strategy enables direct optimization of experimental properties in inverse design.
• The approach shows significant potential for designing diverse organic fluorescent materials with tailored optical characteristics.
• Validation through newly synthesized molecules confirms the framework's efficacy.