Generative Model for Synthesizing Ionizable Lipids: A Monte Carlo Tree Search Approach

TL;DR

This paper tackles the challenge of designing synthesizable ionizable lipids for lipid nanoparticles used in mRNA delivery by introducing a policy-network guided Monte Carlo Tree Search (MCTS) framework. It builds a synthetically accessible lipid building-block dataset, develops lipid-likeness and ionizability predictors, and applies a template-based reaction model within an MCTS guided by a neural policy to generate candidate lipids with feasible synthesis routes. Empirical results show that guided MCTS substantially improves the ionizable-lipid yield (approximately 0.73–0.80) compared with a SyntheMol baseline (≈0.27) and random generation (≈0.15), while maintaining reasonable synthetic-feasibility metrics; retrosynthesis analyses reveal limitations due to dataset-template mismatches, highlighting areas for further development. Overall, the work advances synthesis-aware molecular design for lipid nanoparticles and provides a practical pathway toward rapid discovery of ionizable lipids, with clear directions for experimental validation and template expansion.

Abstract

Ionizable lipids are essential in developing lipid nanoparticles (LNPs) for effective messenger RNA (mRNA) delivery. While traditional methods for designing new ionizable lipids are typically time-consuming, deep generative models have emerged as a powerful solution, significantly accelerating the molecular discovery process. However, a practical challenge arises as the molecular structures generated can often be difficult or infeasible to synthesize. This project explores Monte Carlo tree search (MCTS)-based generative models for synthesizable ionizable lipids. Leveraging a synthetically accessible lipid building block dataset and two specialized predictors to guide the search through chemical space, we introduce a policy network guided MCTS generative model capable of producing new ionizable lipids with available synthesis pathways.

Figures (9)

• Figure 1: Roadmap of lipid building block datasets construction. The upper flowchart illustrates the process of filtering ionizable lipid head building block dataset. The lower flowchart illustrates the process of constructing lipid tail building block dataset.
• Figure 2: Workflow of policy network training.
• Figure 3: Policy network guided Monte Carlo tree search for lipid generation.
• Figure 4: Ionizable lipid rate of the generated products during guided MCTS simulations.
• Figure 5: Selected examples of lipid head building blocks.