Polymer Design via SHAP and Bayesian Machine Learning Optimizes pDNA and CRISPR Ribonucleoprotein Delivery

Abstract

We present the facile synthesis of a clickable polymer library with systematic variations in length, binary composition, pKa, and hydrophobicity (clogP) to optimize intracellular pDNA and CRISPR-Cas9 ribonucleoprotein (RNP) performance. We couple physiochemical characterization and machine learning to interpret quantitative structure-property relationships within the combinatorial design space. For the first time, we reveal unexpected disparate design parameters for nucleic acid carriers; via explainable machine learning on 432 formulations, we discover that lower polymer pKa and higher percentages of benzimidazole ethanethiol enhance pDNA delivery, yet polymer length and captamine cation identity improve RNP delivery. Closed-loop Bayesian optimization of 552 formulation ratios further enhances in vitro performance. The top three polymers yield higher signal and stable transgene expression over 20-days in vivo, and a 1.7-fold enhancement over controls. Our facile coupling of synthesis, characterization, and machine analysis provide powerful tools to quantitate performance parameters accelerating next generation vehicles for nucleic acid medicines.

Data-Driven Design of Polymer-Based Biomaterials: High-throughput Simulation, Experimentation, and Machine Learning

“…Therefore, there is considerable incentive to judiciously select data that crucially advance design goals. Indeed, Bayesian optimization has assisted in efficiently navigating complex chemical spaces to optimize polymer properties in diverse contexts. ,− Our experiences with this strategy mostly derive from collaboration with Gormley and co-workers on designing thermostabilized protein–polymer hybrid systems. ,, In this context, our task was to identify copolymer compositions and polymerization ranges that would maximize retention of enzymatic activity post thermal stressing relative to the unstressed enzyme in its native state. Data for a representative design campaign involving one enzyme, lipase, is shown in Figure .…”

Self-Driving Laboratories: A Paradigm Shift in Nanomedicine Development

“…This process will repeat until a predetermined experimental budget is exhausted. Indeed, AL strategies such as Bayesian optimization [24][25][26] and variance-based sampling 23 have recently been applied in the context of closed-loop design of nanomaterials, including inorganic nanoparticles, 27,28 polymeric nanoparticles encapsulating nucleic acids, 29 and polymer-protein hybrids. 30 Few-shot learning for accurate prediction of nanomedicines for novel active agents.…”

Re-envisioning the Design of Nanomedicines: Harnessing Automation and Artificial Intelligence

“…Using a combination of active ML (e.g., Bayesian optimization) and automated synthesis, candidate copolymers could be iteratively tested, and the data could be looped back into the workflow to determine important chemical features and identify thermostable polymer-protein hybrids. Likewise, Dalal et al recently used a Bayesian optimizer to identify and optimize chemical and physical characteristics of polymer formulations for in vitro nucleic acid delivery [98]. Such studies show how ML workflows can perform closed-loop predictions to gain important QSAR insights for advancing formulation development.…”