Machine-guided design of chaperone-mimetic polymeric carriers for ribonucleoprotein delivery - PROJECT SUMMARY In this Trailblazer R21, our interdisciplinary team [polymers, machine learning (ML), and hematopoietic stem and progenitor cells (HSPCs)] will innovate ML-guided polymer discovery pipelines for intracellular delivery of ribonucleoproteins (RNPs). Chaperone-mimetic polymers developed in this project will augment the safety and site-specificity of genome editing by counteracting intracellular RNP misfolding and performing conforma- tional quality control. Transfer learning frameworks innovated in this project will streamline the discovery of pol- ymeric RNP carriers in hard-to-expand data-scarce cells such as HSPCs that are otherwise recalcitrant to ML- guided experimentation. RNPs are expensive to express and purify. Sadly, electroporation consumes substantial amounts of RNPs, exacerbating already high RNP costs. Polymers are not only affordable to manufacture but also load RNPs efficiently, thereby lowering the costs of RNP therapeutics. However, heuristic optimization of polymer lengths, architectures, or compositions is inefficient and experimentally onerous. We will pioneer ML frameworks that address 2 conceptual blind spots in RNP delivery: conformational quality control of intracellular RNPs (misfolded RNPs trigger deleterious edits) and tailoring polymers for data-scarce HSPCs. Protein payloads such as RNPs unfold during endosomal translocation but they must recover conforma- tional integrity in the cytosol. Otherwise, misfolded RNPs will fail to discriminate between target and off-target DNA sequences, jeopardizing the safety of genome editing. In Aim 1, we will engineer chaperone-mimetic polymers that refold RNPs back into catalytically competent conformations, augmenting the precision and effi- ciency of genome editing. During RNP refolding, polymers must promote peptide self-sorting while preventing aggregation. This requires careful modulation of polymer cationicity and hydrophobicity. Current tools fail to ex- plore vast design spaces along information-efficient trajectories. In contrast, Bayesian optimization (BO) will rapidly identify polymeric chaperones-cum-nanocarriers of optimized hydrophobicity and cationicity. RNP delivery to HSPCs has transformed the therapeutic landscape for hemoglobinopathies but electro- poration-based therapeutics such as exa-cel are prohibitively expensive ($2M per patient). Polymers load RNPs efficiently, consuming fewer RNPs per dose than electroporation and dramatically improving affordability. How- ever, polymers have met with limited success in delivering RNPs to HSPCs. Importantly, HSPCs are challenging to expand into large cell populations, which makes them ill-suited for testing large polymer libraries via data- intensive ML approaches. If we start with 1 million HSPCs, only 3 polymer–RNP complexes can be tested even in miniaturized experimental set-ups; a 150-strong polymer library will require 50–100 million HSPCs, which is experimentally onerous. Instead, in Aim 2, we will transfer information from data-rich cellular domains to tailor polymers for data-scarce HSPCs via transfer learning. This TrailBlazer R21 will alleviate the financial burdens of RNPs by developing an innovative ML-driven blueprint to unlock the potential of polymeric nanocarriers.
