DesignMaster is a diffusion-based generative framework for multi-conditional PROTAC linker design. The model explicitly incorporates linker length and physicochemical properties as controllable conditioning signals, enabling fine-grained and constraint-aware generation of PROTAC molecules.
DesignMaster employs an E(3)-equivariant graph Transformer to model three-dimensional molecular structures and introduces a gated multi-condition fusion module that injects conditional signals throughout the diffusion denoising process. This design allows the model to generate PROTAC linkers that satisfy both structural compatibility and physicochemical constraints.
Experiments on PROTAC-DB 2.0 and PROTAC-DB 3.0 demonstrate that DesignMaster achieves state-of-the-art performance in terms of validity and recovery, while also producing structures with improved geometric fidelity.
The training datasets, pretrained checkpoints, and generated results are publicly available at Zenodo:
Zenodo repository
https://zenodo.org/records/18883030
The archive includes:
- Training datasets
- PROTAC-DB 2.0
- PROTAC-DB 3.0
- Pretrained model checkpoints
- Generated PROTAC samples
Please download and extract the files before running training or inference.