(182y) A Data Analytics Approach for Rational Design of Nanomedicines with Programmable Drug Release

Nanoparticle-based drug delivery vehicles can improve the potency of antimicrobial drugs by enhancing intracellular localization and targeting the delivery of their cargo to a pathogen’s intracellular niche within host cells. Polyanhydride nanoparticles show passive targeting and payload stabilization properties that make them uniquely suited for antimicrobial delivery for intracellular infections. A key property of these nanomedicines is their ability to control the release kinetics of their encapsulated payload. Release kinetics from these nanoparticles is challenging to model from first principles due to interactions between polymer, drug, and nanoparticle properties. Further, the near-infinite combinations of polymers and drug candidates yields an immense parameter space of properties, which challenges the feasibility of traditional screening approaches.

Hierarchical modeling represents a more efficient approach to design these systems, wherein key polymer and drug descriptors are identified and correlated to release behavior. To date, most modeling of nanomedicine drug release has focused either on linear dimensionality reduction or artificial neural network methods. The former typically employs principal component analysis (PCA) and regression through partial least squares, which provides insights into relationships between formulations and variables but has limited capacity to capture nonlinear behavior inherent to nanomedicines. By contrast, artificial neural network “black box” models can capture non-linear behavior but obscure interpretation of the structure of the model and dataspace. A modeling methodology capable of capturing non-linear behavior while preserving relationships between formulations is an optimal, but elusive, solution.

In this work, a nanoparticle library of 68 formulations was synthesized and release kinetics of various drugs were measured. Polyanhydride copolymers with varying composition were synthesized from 1,8-bis(p-carboxyphenoxy)-3,6-dioxaoctane (CPTEG), 1,6-bis(p-carboxyphenoxy)hexane (CPH), and sebacic acid (SA) monomers and used to encapsulate four antibiotic payloads: doxycycline, rifampicin, chloramphenicol, and pyrazinamide. Release kinetics were parameterized to represent burst release, sustained release, and encapsulation efficiency of each formulation.

PCA and variable importance projection were used to identify key polymer, drug, and nanoparticle descriptors correlated with release kinetics behavior. PCA demonstrated that drug release properties were influenced by both polymer and drug properties, as expected. However, PCA was unable to adequately model release behavior, likely due to the aforementioned non-linear dynamics inherent to nanomedicine systems. A graph theory analysis was used to map nanoformulation behavior in multilinear space, then project the underlying higher-dimensional manifold onto a 2-D plane. This analysis provided a map of nanoformulations that simultaneously captures non-linear behavior while preserving nearest-neighbor relationships between nanoformulations. Accordingly, connectivity contained in this map allows interrogation of design pathways between formulations, enabling selection of alternate drug payloads or polymer chemistry based on similarity of release behavior.

Graph theory was used to generate predictive multilinear models of nanoparticle release behavior, which were cross validated to ensure a balance between robustness and accuracy. These models were fairly accurate, with R² values ranging between 70.0% and 75.5%. To test the predictive capabilities of these models, eight new formulations were generated, encapsulating two antibiotics, meropenem and ceftazidime, not included in the training data set. The predictive models showed good agreement with experimental results, supporting the use of this methodology to virtually explore novel nanomedicine formulations with polymer and drug pairs. In these ways, this multilinear modeling approach provides the first steps towards development of a framework that can be used to rationally design nanomedicine formulations by selecting the appropriate carrier for a drug payload to program desirable release kinetics profiles.