MolQuery: Prediction of Lipid Synthesizability Using Active Learning

In the field of molecular design, Generative Artificial Intelligence (GenAI) has the potential to create extensive synthetic data sets encompassing a wide range of chemical properties. However, the practical application of these data sets is often constrained by the synthesizability of the molecules within. To address this, it is essential to develop a robust platform for assessing synthesizability, which is crucial for constructing effective GenAI-based models for molecular systems. Here, we introduces MolQuery, a comprehensive pipeline that integrates active learning (AL) to improve the accuracy of chemical synthesizability predictions for lipid molecules designed for mRNA delivery via lipid nanoparticles (LNPs). By leveraging AL, MolQuery efficiently trains machine learning models using small data sets which greatly improves upon current solutions for this tasks. Our results demonstrate that MolQuery produces highly accurate predictions of lipid synthesizability, making it a valuable tool for filtering synthetic LNP data sets.