A Deep Generative Model for the Design of Synthesizable Ionizable Lipids

TL;DR

The paper tackles the challenge of designing synthesizable ionizable lipids for lipid nanoparticle delivery by developing a deep generative model based on Synthesis-DAGs, extended with a domain-specific ionizable lipid dataset and a more capable reaction predictor (Chemformer). It builds a large ionizable lipid synthesis dataset from synthetically accessible building blocks, trains property and ionizability predictors, and integrates AGILE to optimize transfection efficiency in HeLa cells. The approach achieves an ionizable lipid generation rate of 83.4% and demonstrates that DAG-based synthesis path representations yield superior generation quality compared to linear representations, with potential for discovering lipids that enhance mRNA delivery. Limitations include predictor validity and the need for experimental wet-lab validation to confirm synthesis feasibility and transfection performance, which the authors propose as future work to close the loop with experimental validation.

Abstract

Lipid nanoparticles (LNPs) are vital in modern biomedicine, enabling the effective delivery of mRNA for vaccines and therapies by protecting it from rapid degradation. Among the components of LNPs, ionizable lipids play a key role in RNA protection and facilitate its delivery into the cytoplasm. However, designing ionizable lipids is complex. Deep generative models can accelerate this process and explore a larger candidate space compared to traditional methods. Due to the structural differences between lipids and small molecules, existing generative models used for small molecule generation are unsuitable for lipid generation. To address this, we developed a deep generative model specifically tailored for the discovery of ionizable lipids. Our model generates novel ionizable lipid structures and provides synthesis paths using synthetically accessible building blocks, addressing synthesizability. This advancement holds promise for streamlining the development of lipid-based delivery systems, potentially accelerating the deployment of new therapeutic agents, including mRNA vaccines and gene therapies.

Figures (4)

• Figure 1: An Example of Ionizable Lipid Structure.
• Figure 2: Examples of Predictions from the Original and Improved Reaction Predictors. The original reaction predictor, Molecular Transformer, fails to accurately predict reactions between lipid heads and tails, with errors indicated by blue boxes. In contrast, Chemformer successfully predicts the correct reactions.
• Figure 3: Examples of Generated Ionizable Lipids and Their Synthesis Paths. The synthesis paths show the building blocks and intermediate products. Our model can generate ionizable lipids with one to three tails.
• Figure 4: Ionizable Lipids mRNA Transfection Efficiency in HeLa Cells.