Fig. 5: Structure analysis of ionizable lipids informed by AI models.

a Ionizable lipid pattern and tail types for analysis. b–d Heatmap of positive rates for specific tail types. All types of tails and heads containing hydroxyl were combined to form ionizable lipids in an exhaustive manner. Their mRNA delivery efficiency and apparent pKa were predicted with Model 1. For each type of tail, the number of resulting lipids containing the tail and among them the number of positive lipids (efficiency higher than the standard MC3 formulation and pKa between 6.0 and 7.0) were used to calculate the positive rate. The structure-activity relationship is shown as the influence of tail length and linker position (b) tail length and branch length (c), and the branch length and linker position (d) on the positive rate. AI, artificial intelligence.