Reproduce “A Deep Generative Model for the Design of Synthesizable Ionizable Lipids” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Yuxuan Ou, Jingyi Zhao, Austin Tripp, Morteza Rasoulianboroujeni, José Miguel Hernández-Lobato
A deep generative model was developed to design ionizable lipids for mRNA delivery, generating both lipid structures and their synthesis pathways from commercially available building blocks. This approach yielded an 83.4% ionizable lipid generation rate and a Fréchet ChemNet Distance of 3.797, demonstrating high quality and synthetic accessibility in the generated compounds.
Key insights
- Integrating the generation of synthesis pathways directly with novel molecular structures from accessible building blocks is an effective strategy to address the synthesizability challenge in computational molecular design.
- The use of DAG representations for synthesis pathways, coupled with a high-performance reaction predictor like Chemformer, significantly improves the efficiency and chemical quality of generated lipids compared to linear path representations or less accurate predictors.
- Iterative fine-tuning of the generative model, guided by property prediction models, enables the optimization of generated lipid structures towards specific biological functions, such as high mRNA transfection efficiency.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.