Related Experiment Video
Updated: Jun 23, 2026

Curation of Computational Chemical Libraries Demonstrated with Alpha-Amino Acids
Published on: April 13, 2022
Modular Framework for 3D Molecular Generation in Computational Chemistry Applications
Thanapat Worakul1, Mohammed Azzouzi2, Matthew D Wodrich1,3
1Laboratory for Computational Molecular Design, Institute of Chemical Sciences and Engineering, Ecole Polytechnique Fédérale de Lausanne (EPFL), 1015 Lausanne, Switzerland.
MolCraftDiffusion is a new platform for 3D molecular generative models, addressing computational cost and fragmentation challenges. It enables efficient development, evaluation, and deployment of 3D molecular diffusion models for computational chemistry tasks.
Area of Science:
- Computational Chemistry
- Artificial Intelligence
- Molecular Modeling
Background:
- Three-dimensional (3D) molecular generative models are crucial for drug discovery and materials science, but their adoption is hindered by high computational costs and a lack of standardized platforms.
- Existing 3D generative models are fragmented, with scattered implementations and evaluation protocols, limiting reproducibility and comparison.
Purpose of the Study:
- To introduce MolCraftDiffusion, a modular and extensible platform for developing, evaluating, and deploying 3D molecular diffusion models.
- To overcome the limitations of computational cost and fragmentation in 3D molecular generation.
Main Methods:
- Developed a layered architecture decoupling core training logic from model definitions and task implementations for enhanced modularity and extensibility.
- Implemented curriculum learning for progressive training on diverse 3D molecular datasets, creating a pretrained diffusion model to reduce downstream training costs.
- Integrated modular guidance mechanisms, including molecular inpainting/outpainting for structure control and property-conditioned generation.
Main Results:
- Demonstrated the platform's capabilities across computational chemistry tasks like virtual library construction and inverse molecular design.
- Successfully integrated three distinct existing models (TABASCO, ADiT, ShEPhERD) without core codebase modifications, showcasing extensibility.
- Provided a pretrained diffusion model and a comprehensive framework for reproducible research and development.
Conclusions:
- MolCraftDiffusion offers a standardized, efficient, and extensible solution for 3D molecular generative modeling.
- The platform facilitates the development and application of advanced molecular design tools, accelerating computational chemistry research.
- The modular design and pretrained models lower the barrier to entry for researchers in the field.
Related Concept Videos
Molecular Models
Predicting Molecular Geometry
Molecular Geometry and Dipole Moments
Molecular Shapes
VSEPR Theory
Newman Projections
The organic molecules rotate across the single bonds leading to numerous temporary three-dimensional structures of varying energy known as conformers.

