Leveraging AI to Improve Safety and Potency of Lung-Specific Lipid Nanoparticles

Lipid nanoparticles (LNPs) have revolutionized RNA-based therapeutics. However, ensuring safe and effective delivery of mRNA to extra-hepatic tissues, such as the lungs, remains a significant hurdle largely due to challenges with cell-specific targeting and systemic toxicity. To address these challenges, Mana.bio has developed an machine learning (ML) platform capable of predicting critical LNP features required for specific and tolerable systemic mRNA delivery. The power and utility of this platform are demonstrated through Mana’s lung delivery program, which has leverages ML to model physicochemical properties, in-vivo targeting, and systemic safety toward unlocking the potential mRNA therapeutics for lung diseases.

Mana.bio’s platform integrates over 80,000 proprietary data points and 45,000 publicly available data points, including results from in-vitro, ex-vivo, and in-vivo models. Leveraging advanced ML capabilities, we can accurately predict physicochemical properties—such as size, charge, and encapsulation efficiency, in-vivo lung targeting, and key toxicity parameters—such as cytokine secretion, liver enzyme elevations, and coagulation risks—while screening, in-silico, tens of millions of potential LNP formulations. The model ranks LNP candidates based on physicochemical properties, activity probabiility, and safety profile, enabling rapid identification of optimized formulations for specific therapeutic applications.

The platform initially identified MB-LNP-01, an intravenously administered LNP candidate with highly specific lung delivery. Subsequent optimization efforts, incorporating only in-silico and in-vitro data, led to MB-LNP-02, a second-generation candidate with 100-fold increased potency. These findings highlight the platform’s ability to rapidly enhance therapeutic performance while reducing development timelines. Towards improving the potential therapeutic index of MB-LNP-02, Mana.bio’s safety prediction model identified subsequent novel LNP candidates by optimizing for critical toxicity assessments such as direct toxicity, immunotoxicity, hepatotoxicity, and coagulation. Through iterative refinement, MB-LNP-05 was engineered and validated in C57bl/6 mice, demonstrating above 100-fold reduction in cytokine elevation and a 100% improvement in tolerable dose with no off-target effects in key organs. Importantly, MB-LNP-05 exhibited enhanced lung-specific mRNA expression compared to MB-LNP-02, measured 6 hours after an IV dose of 0.65 mg/kg and 1.2 mg/kg fluc mRNA. Non-Human Primate (NHP) studies further confirmed these improvements, showing a 30-fold reduction in inflammatory cytokines, including IL-6, TNF-α, and IP-10, even at double the dose of MB-LNP-02. ALT and AST levels remained normal, with no acute liver damage observed at the maximum tested dose.

This work underscores the transformative potential of integrating ML-driven modeling with experimental validation to accelerate drug development and unlock the potential of RNA medicines.