Shruti Patel

Evaluation of Vector Representations of Lipid Nanoparticles in Cheminformatic Predictions of Transfection Efficiency

Started Sept. 11, 2024

Abstract or project description

Lipid nanoparticles (LNPs) are a revolutionary form of drug delivery for RNA-based therapeutics as they are difficult to degrade and can efficiently transport its contents to farther target cells. Optimizing LNP formulations is essential for improving therapies, yet the best way to computationally represent these formulations for predictive modeling remains unexplored. This paper analyzes different strategies for constructing the formulation vector of LNPs to evaluate their impact on the accuracy of predicting transfection efficiency. Specifically, three approaches were examined: (1) fully describing all lipid components using molecular descriptors, (2) fully describing only the cationic lipid while incorporating molar ratios for other components, and (3) fully describing both the ionizable and helper lipids while using molar ratios for rest. These machine learning models were trained using each formulation representations, and the results revealed minimal differences in predictive accuracy. The results suggest that the most important structures to be considered are the cationic and helper lipids, and including molecular descriptors for the PEGylated lipid and cholesterol may not be necessary for predicting delivery efficiency, as it may cause excess noise in the neural network. This can streamline LNP formulation research, which generally takes years of testing to design specific LNPs. By identifying effective ways to represent LNP formulations, this project contributes to optimizing drug delivery systems.