This repository contains the code for running the experiments of the thesis "Guiding Diffusion with Logical Constraints: Molecular Graph Generation under Lipinski's Rules", developed at Universitat Politècnica de Catalunya (UPC), Facultat d’Informàtica de Barcelona (FIB), jointly with the University of Padua, Department of Mathematics, as part of the Master’s in Data Science Double Degree Programme. The thesis can be freely accessed at Thesis and Dissertation Padua Archive.
This project is adapted from DiGress, developed by Clement Vignac, Igor Krawczuk, and Antoine Siraudin, and distributed under the MIT License.
Extensions and modifications © 2025 Emma Meneghini.
The repository combines adapted components from the DiGress guidance branch with original contributions developed for this thesis:
-
./src/– Adapted code from DiGress, with corrections and substantial extensions.
Thedatasets,guidance, andmodelssubfolders were expanded with new functionality, and several issues from the original implementation were addressed. -
./configs/– Configurations for running the experiments, with significant additions in theexperiment/subfolder. -
./evaluation_scripts/– Entirely new code for extracting detailed statistics from the generated molecule sets.
The setups and requirements provided in this repository should be used to build the environment. However, the compilation and execution procedures remain the same as those described in the original DiGress repository.
This work was carried out under the supervision of Dr. Sergi Abadal Cavallé (UPC-FIB) and Dr. Nicolò Navarin (University of Padua, Department of Mathematics).
This repository is distributed under the MIT License. See LICENSE for details.