Automation and data-driven design of polymer therapeutics

Abstract

Polymers are uniquely suited for drug delivery and biomaterial applications due to tunable structural parameters such as length, composition, architecture, and valency. To facilitate designs, researchers may explore combinatorial libraries in a high throughput fashion to correlate structure to function. However, traditional polymerization reactions including controlled living radical polymerization (CLRP) and ring-opening polymerization (ROP) require inert reaction conditions and extensive expertise to implement. With the advent of air-tolerance and automation, several polymerization techniques are now compatible with well plates and can be carried out at the benchtop, making high throughput synthesis and high throughput screening (HTS) possible. To avoid HTS pitfalls often described as "fishing expeditions," it is crucial to employ intelligent and big data approaches to maximize experimental efficiency. This is where the disruptive technologies of machine learning (ML) and artificial intelligence (AI) will likely play a role. In fact, ML and AI are already impacting small molecule drug discovery and showing signs of emerging in drug delivery. In this review, we present state-of-the-art research in drug delivery, gene delivery, antimicrobial polymers, and bioactive polymers alongside data-driven developments in drug design and organic synthesis. From this insight, important lessons are revealed for the polymer therapeutics community including the value of a closed loop design-build-test-learn workflow. This is an exciting time as researchers will gain the ability to fully explore the polymer structural landscape and establish quantitative structure-property relationships (QSPRs) with biological significance.

Keywords: Artificial intelligence; Automation; Drug delivery; Gene delivery; High throughput screening; Machine learning; Polymer chemistry.

Fig. 1.

PubMed hits for peer-reviewed publications involving ML, AI, pharmaceuticals, and drug delivery over the period from 2000-2019. (A) Number of publications from 2000-2019 containing “machine learning,” “artificial intelligence,” and “pharmaceutical.” (B) Number of publications from 2000-2019 containing “machine learning,” “artificial intelligence,” and “drug delivery.”

Fig. 2.

Various bioactive polymer architectures are possible in high throughput using cp-DIBAC. Linear functional polymers can be synthesized by polymerizing cp-DIBAC monomer (M1) onto the polymer backbone. End-functionalized 2-arm, 3-arm, and 4-arm polymers can also be synthesized by utilizing respective RAFT agents (R2, R3, and R4). Deprotection is conducted at 290-350 nm prior to clicking ligand. Reprinted with permission from the American Chemical Society. Copyright 2019 American Chemical Society [20].

Fig. 3.

Schematic of automated polymer synthesis workflow incorporating liquid handling robotics and open-air Enz-RAFT and PET-RAFT. User inputs such as DP, polymer composition, and CTA type are supplied to a Python script which produces pipetting sequences, concentrations, dispensing volumes, and process information. The Hamilton MLSTARlet liquid handling robot carries out the open-air chemistry directly in 96-well plates. This process is compatible with homopolymers, random heteropolymers, and block copolymers [10]. Reproduced with permission from Wiley-VCH Verlag GmbH & Co. KGaA. License can be found online (https://creativecommons.org/licenses/by/4.0/).

Fig. 4.

Automated synthesis of functionalized polymers. The user first designs the polymer library by specifying DP, composition, and functionalization information. Reagents are loaded by the user and then the liquid handling robot carries out dispensing steps. Post-polymerization modification was validated for strain-promoted azide-alkyne cycloaddition (SPAAC) [10]. Reproduced with permission from Wiley-VCH Verlag GmbH & Co. KGaA. License can be found online (https://creativecommons.org/licenses/by/4.0/).

Fig. 5.

Evolution of analytical chromatography from the 1990s to the present day. Over time, the limits of column size, particle size, and run times have dramatically decreased. Higher pressure limits and a wider pH range are also possible [132]. Reproduced with permission from Elsevier.

Fig. 6.

ML approaches in the context of polymer chemistry. In polymer chemistry, relevant input parameters that can be controlled via the selected synthetic approach are DP, polymer composition, monomer arrangement, polymer architecture, and valency. PCA and BO aim to determine which parameters most contribute to the variance in data while RF, SVM, GP, and ANN can be utilized in property prediction.

Fig. 7.

Representation block organization of polymers alongside a heat map of MIC for each. This enabled identification of hit polymer candidates that exhibited high bioactivity and warrant closer consideration [161]. Reproduced with permission from Wiley-VCH Verlag GmbH & Co. KGaA.

Fig. 8.

Characterizing toxicity and screening antimicrobial polymer candidates after automated synthesis. Along with base monomer PDMAEMA, co-monomers were chosen and synthesized into the polymer backbone at varying mol%. There was shown to be minimal hemolytic activity for all polymers. PPGMA was identified to have the most bioactivity overall as the most hydrophilic monomer [25]. Reproduced with permission from Wiley-VCH Verlag GmbH & Co. KGaA.

Fig. 9.

The effect of polymer concentration and degree of branching on transfection efficacy, expression, and viability. A rapid combinatorial synthesis approach allowed further study of these conditions prior to additional screening experiments. Reprinted and adapted with permission from the American Chemical Society. Copyright 2019 American Chemical Society [174].

Fig. 10.

Preparation and mechanism of ASD interaction with drug. ASDs shown are prepared via spray drying to form a polymer and drug solid mixture. Polymer-drug interactions, typically through non-covalent interactions, enable API supersaturation whereby precipitation or crystallization of API is prevented over a long period of time. Reprinted with permission from the American Chemical Society. Copyright 2015 American Chemical Society (https://pubs.acs.org/doi/10.1021/acsbiomaterials.5b00234) [176]. Requests for permissions should be directed to ACS.

Fig. 11.

Comparison of various modified HPMCAS polymers for supersaturation of probucol, danazol, and phenytoin at different drug wt% (10, 25, and 50 wt%). CEA-modified HPMCAS performed best in dissolving probucol, likely due to hydrogen bonding and hydrophobic interactions. Reprinted with permission from the American Chemical Society. Copyright 2015 American Chemical Society (https://pubs.acs.org/doi/10.1021/acsbiomaterials.5b00234) [176]. Requests for permissions should be directed to ACS.

Fig. 12.

Effect of polymer DP and architecture on binding concanavalin A (ConA) via a lectin binding assay. The percentage of bound ConA is displayed for various DP linear, 3-arm, and 4-arm polymers. 3-arm polymers performed best, especially at lower DPs, followed by linear and then 4-arm polymers. This illustrates the utility of the combinatorial library approach in determining structure-activity relationships [9]. Reproduced with permission from Wiley-VCH Verlag GmbH & Co. KGaA.

Fig. 13.

ML approach to identify polymers that have high thermal conductivity. Bayesian molecular design was initially conducted to train a model to predict secondary properties of T_g and T_m. Transfer learning was then completed with limited data on thermal conductivity to generate predictions which were verified through the synthesis of three polymers [250]. Reproduced with permission from Springer Nature. License can be found online (https://creativecommons.org/licenses/by/4.0/).

Fig. 14.

Robotic and data-driven design system for CASP. (A) Comparison of this work to previously published work (gray bars denote areas in which automation has been achieved). This work has demonstrated automation in every aspect of the organic synthesis process except recipe formulation which requires some user input (e.g. stoichiometry and confirming reaction conditions). (B) ML approach that utilizes chemical reaction data found in the Reaxys database. (C) Image of the six-axis robotic manipulator along with modular setup of the working area [259]. Reproduced with permission from the American Association for the Advancement of Science.