KFO-Gen: A conditional variational autoencoder trained on KFO-Atlas for generative aroma design

KFO-Gen is a generative aroma design model based on a conditional variational autoencoder (CVAE) trained on KFO-Atlas. The model learns the latent structure of key food odorant (KFO) combinations and their relative proportions, enabling the computational generation of novel aroma formulations conditioned on target aroma categories.

This model accompanies the publication “Molecular Atlas of Key Food Odorants Reveals Mixture-Level Organization and Enables Generative Aroma Design,” published in Advanced Science (2026).

The code and model provided in this repository are licensed under the Creative Commons Attribution-NonCommercial 4.0 International (CC BY-NC 4.0) License. They may be freely used for academic research and other non-commercial purposes with appropriate attribution. For commercial use, please contact the authors to obtain permission.