DyeLeS: a web platform for predicting and classifying fluorescence properties of bioactive molecules

Area of Science:

• Computational chemistry
• Medicinal chemistry
• Drug discovery

Background:

• Fluorescent drug molecules are crucial for real-time tracking in biomedical research and precision medicine.
• Traditional fluorescent drug discovery methods are inefficient and resource-intensive due to reliance on trial-and-error.

Purpose of the Study:

• To develop a computational framework, DyeLeS (Dye-Likeness Scoring), for rapid evaluation of molecular fluorescence potential.
• To create FluoBioDB, the first public library of fluorescent bioactive compounds.

Main Methods:

• DyeLeS uses a Naive Bayes-inspired algorithm for fluorescence classification (AUC = 0.995) and a LightGBM model for predicting photophysical properties (R² = 0.88 for λabs).
• The study curated a dataset of fluorescent and non-fluorescent compounds to train the models.
• FluoBioDB was constructed using DyeLeS, containing 32,865 diverse fluorescent bioactive molecules.

Main Results:

• DyeLeS accurately predicts fluorescence potential and key photophysical properties.
• FluoBioDB comprises a large, diverse set of fluorescent compounds, including kinase inhibitors and GPCR modulators.
• Analysis revealed common structural features in fluorescent compounds: polycyclic conjugated frameworks, donor-acceptor structures, and rigid planar cores.

Conclusions:

• DyeLeS provides a robust computational framework to accelerate fluorescent drug discovery.
• FluoBioDB serves as a valuable resource for identifying and optimizing fluorescent drug candidates.
• The developed tools and resource support advancements in bioimaging, targeted therapy, and theranostics.