Optimization of Mixed Micelles Based on Oppositely Charged Block Copolymers by Machine Learning for Application in Gene Delivery

The COVID-19 mRNA vaccines represent a milestone in developing non-viral gene carriers, and their success highlights the crucial need for continued research in this field to address further challenges. Polymer-based delivery systems are particularly promising due to their versatile chemical structure and convenient adaptability, but struggle with the toxicity-efficiency dilemma. Introducing anionic, hydrophilic, or "stealth" functionalities represents a promising approach to overcome this dilemma in gene delivery. Here, two sets of diblock terpolymers are created comprising hydrophobic poly(n-butyl acrylate) (PnBA), a copolymer segment made of hydrophilic 4-acryloylmorpholine (NAM), and either the cationic 3-guanidinopropyl acrylamide (GPAm) or the 2-carboxyethyl acrylamide (CEAm), which is negatively charged at neutral conditions. These oppositely charged sets of diblocks are co-assembled in different ratios to form mixed micelles. Since this experimental design enables countless mixing possibilities, a machine learning approach is applied to identify an optimal GPAm/CEAm ratio for achieving high transfection efficiency and cell viability with little resource expenses. After two runs, an optimal ratio to overcome the toxicity-efficiency dilemma is identified. The results highlight the remarkable potential of integrating machine learning into polymer chemistry to effectively tackle the enormous number of conceivable combinations for identifying novel and powerful gene transporters. Mixed micelles that contain the positively charged guanidinium-, negatively charged carboxy-, and hydrophilic groups in their shell are assembled by mixing cationic and anionic diblock terpolymers in varying ratios. A machine learning model helps to find the optimal cationic to anionic ratio for the mixed micelle composition, which shows high transfection and viability, overcoming the toxicity-efficiency dilemma.image