.png){kind=icon}
.png){kind=icon}
Return
Semantic substructure guided multiple objective molecular generation with discrete diffusion probabilistic model
.png){kind=icon}
PRE
AI
DOI：10.1016/j.neucom.2025.131086.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Deep learning methods have paved the highway for drug discovery in past few years, it is still a big challenge that designed molecules satisfy multiple objective constraints. Traditional generative adversarial networks suffer severe mode collapse problems and Transformer based large language models require relative large amount of data to train or fine-tune. Denoising diffusion probabilistic models have achieved huge success in image generation fields, yet it remains an open challenge to explore diffusion based language models in drug discovery. Herein, we bridge the molecule prior distribution with the desired bioactive molecular distribution and propose SGDiff, Substructure Guided Diffusion model, which can iteratively inject the sequence-based conditional information into the denoising process to guide the generation process without additional reinforcement learning fine-tuning process. We conduct sufficient discussion about the Transformer based language models and diffusion based language models. SGDiff can achieve the highest conditional success rate in multiple objective generation tasks and can maintain high diversity from conditional guidance. Besides, SGDiff has a great interpretation during sampling. Several explorations have been conducted to detail understanding the mechanism of the guidance paradigm, which can be better used for further multiple objective constraints molecular generation.
Keywords:
diffusion models
drug discovery
molecular generation
conditional guidance
substructure guided diffusion
Journal
IF:
6.5
Papers:
2.5W
Citations:
6.5W