Inverse molecular design with diffusion models holds great potential for
advancements in material and drug discovery. Despite success in unconditional
molecule generation, integrating multiple properties such as synthetic score
and gas permeability as condition constraints into diffusion models remains
unexplored. We introduce multi-conditional diffusion guidance. The proposed
Transformer-based denoising model has a condition encoder that learns the
representations of numerical and categorical conditions. The denoising model,
consisting of a structure encoder-decoder, is trained for denoising under the
representation of conditions. The diffusion process becomes graph-dependent to
accurately estimate graph-related noise in molecules, unlike the previous
models that focus solely on the marginal distributions of atoms or bonds. We
extensively validate our model for multi-conditional polymer and small molecule
generation. Results demonstrate our superiority across metrics from
distribution learning to condition control for molecular properties. An inverse
polymer design task for gas separation with feedback from domain experts
further demonstrates its practical utility.

逆向分子设计的多条件扩散模型与转换器编码去噪模型相结合，成功实现了高性能材料和药物发现，并在聚合物和小分子生成任务中展示了优越的分布学习和条件控制表现。