SyntheFluor-RL Data

This dataset accompanies the paper: Sayana, R., Callon, K., Xu, J., Deutsch, J., Chu, S., Zou, J., Janetzko, J., Shivnaraine, R.V., Swanson, K. Generating readily synthesizable small molecule fluorophore scaffolds with reinforcement learning. The 2nd Workshop on Generative AI and Biology, ICML, 2025.The dataset is organized into the following folders: building_block_prep: Intermediate files generated while conducting initial scoring of building blocks for generation. Final input building block file for generation is in synthefluor_generation. chemfluor_data: Chemfluor data files including post-processed files used for training the property prediction models. gaussian_logs: Example Gaussian log files generated in our workflow. Gaussian outputs the results of its calculations (e.x geometry optimization, frequency analysis, energy calculations) as it runs, which we pipe into `.log` files for further analysis. generated_molecule_analysis: intermediate files created to analyze generated molecules and compare with a random set of 10,000 molecules from the REAL space property_prediction_models: Chemprop-RDKit and Chemprop-Morgan models trained to predict PLQY, absorption, or emission properties of molecules. supplementary_figures: UV/HPLC and mass spectrometry analysis of the top three compounds generated by SyntheFluor-RL. supplementary_tables: The supplementary tables referenced in the paper, including model performance of the property prediction models and brightness of the generated molecules. supplementary_software: A custom MATLAB script that was used to quantify mean pixel intensity across fluorescence-only images. synthefluor_generation: Molecules generated by SyntheFluor-RL. This folder additionally includes the building blocks file used as input for generation and the clustering data and analysis of the generated molecules which was used for candidate selection.