AI-Assisted Fluorescence Design

Powered by the Nature Communications (2025) FLSF deep learning model
for intelligent prediction of molecular fluorescence properties, accelerating experimental design and screening.

Model Overview

FLSF (Fluorescence Learning Structure-Feature Model) is a deep neural network-based predictive model for fluorescence performance. By combining molecular structure (SMILES) and solvent effects, it achieves high-accuracy predictions of absorption wavelength, emission wavelength, photoluminescence quantum yield (PLQY), and energy gap.

📊 Dataset: FluoDB and Structural Features

The FLSF model is trained on an integrated fluorescence dataset combining both in-house and public data sources, covering 16 representative molecular backbones, including but not limited to:

  • Fluorophore–solvent combinations: 55,169
  • Distinct fluorescent molecules: 35,528
  • Total optical records: 109,054
  • Includes λabs, λem, ΦPL, εmax, and other multidimensional properties
  • Carbazole series
  • Coumarin and related derivatives
  • Anthracene and Pyrene derivatives
  • Quinoline and Phenoxazine
  • Carbazole–Thiophene conjugated systems
  • Classical dye families such as BODIPY, Rhodamine, and Cyanine

🧠 Model Functions

  • Input: SMILES molecular structure and solvent name
  • Output: abs_pred (absorption peak, nm), emi_pred (emission peak, nm), plqy_pred (PLQY), e_pred (energy gap, eV)
  • Reference: Nature Communications, 2025, DOI: 10.1038/s41467-025-58881-5

Fluorescence Property Prediction

The AI model is running, please wait...