Research

Area I: Machine Learning-accelerated Molecular Design of High-performance Polymers

My first area of research interest is in the machine learning-accelerated molecular design of high-performance polymers. Nowadays, synthetic polymers are used in almost all walks of our life. The spreading of polymer use is connected to their unique properties: low density, low cost, good thermal/electrical insulation properties, high resistance to corrosion, low-energy demanding polymer manufacture, and facile processing into final products. Nevertheless, the development of polymers has traditionally been an experimental-driven, trial-and-error process, guided by experience, intuition, and conceptual insights. This approach is, however, often costly, slow, biased towards certain domains of chemical space, and limited to relatively small-scale studies, which may easily miss promising compounds. In addition, automation of organic molecules and materials design is considerably less developed than that for inorganic materials due to challenges associated with searching the vast design space (on the order of 10⁶⁰-10¹⁰⁰), defined by the almost infinite combinations of the molecular constituent, microstructures, and synthesis conditions.

To overcome this challenge, we formulate a data-driven machine-learning (ML) approach, integrated with high-fidelity molecular dynamics (MD) simulations, for quantitatively predicting the polymer’s thermal (glass transition temperature, Tg) and mechanical properties from its chemical structure and rapid screening of promising candidates for high-temperature polymers. Specifically, we collect a diverse set of nearly 13,000 real polymers from the largest polymer database, PoLyInfo. Among them, 6,923 experimental Tg values are available; while, the remaining 5,690 polymers do not have reported values. We train the deep neural network (DNN) model with 6,923 experimental Tg values using Morgan fingerprint representations of chemical structures for these polymers. Interestingly, the trained DNN model can reasonably predict the unknown Tg values of polymers with distinct molecular structures, in comparison with MD simulations and experimental results. We find that this excellent transferability is attributed to the feature representation of Morgan fingerprints, which carry the chemical connectivity between neighboring repeating units in a polymer and the frequency of occurrence of different chemical substructures. With the validated ML model for high-throughput screening of nearly 1 million hypothetical polymers from a generative ML model, we identify more than 65,000 promising candidates with >200 ^oC, which is 30 times more than existing known high-temperature polymers (~2,000 from PoLyInfo). Our study demonstrates the ability of ML models to generate new compounds and new understanding in the traditionally challenging field of high-temperature polymers by leveraging growing databases of experimental and computational data alongside predictive and generative ML models. We expect this work can address a wide range of scientific questions in computational materials design and synthesis-structure-property relationships for polymeric materials. It will also benefit the broader scientific community and industry, which are interested in developing new types of polymers for sustainable energy solutions, such as polymer membranes for desalination and water treatment.

[1] Yue, Tianle, Jinlong He, Lei Tao, and Ying Li. “High-Throughput Screening and Prediction of High Modulus of Resilience Polymers Using Explainable Machine Learning.” Journal of Chemical Theory and Computation 19, no. 14 (2023), 4641–4653.

[2] Tao, Lei, Jinlong He, Nuwayo Eric Munyaneza, Vikas Varshney, Wei Chen, Guoliang Liu, and Ying Li. “Discovery of multi-functional polyimides through high-throughput screening using explainable machine learning.” Chemical Engineering Journal 465 (2023): 142949.

[3] Yang, Jason, Lei Tao, Jinlong He, Jeffrey R. McCutcheon, and Ying Li. “Machine learning enables interpretable discovery of innovative polymers for gas separation membranes.” Science Advances 8, no. 29 (2022): eabn9545.

[4] Tao, Lei, Vikas Varshney, and Ying Li. “Benchmarking machine learning models for polymer informatics: an example of glass transition temperature.” Journal of Chemical Information and Modeling 61, no. 11 (2021): 5395-5413.

[5] Tao, Lei, Guang Chen, and Ying Li. “Machine learning discovery of high-temperature polymers.” Patterns 2, no. 4 (2021): 100225.

Area II: Cyber-enabled Design of Lipid-Like Theranostic Nanoparticles with Tunable Size and Shape

      My second area of research interest is in the computational design of lipid-like theranostic nanoparticles (NPs) with tunable size and shape for efficiently delivering therapeutics and imaging agents. Liposomes or lipid-like NPs have been widely used as drug delivery vehicles due to their biocompatibility and degradability, including the recently developed COVID-19 mRNA vaccines. Despite the breakthrough in producing vesicles with controlled geometry and surface properties, the current design of liposomes faces challenges in one or more of the following aspects: (1) it remains a challenge to control the size and shape of vesicles within nanometer precision; (2) some existing size-control methods are limited to specific lipid composition without adaptability; (3) liposomes have limited stability during the early phases of drug delivery within blood flow and can leak their contents before arrival at their intracellular target due to membrane pore formation and degradation.

To overcome these challenges, we present a rationally designed lipid-like NP formulation that mimics a liposome’s outer bilayer but can help achieve a uniform size and shape through a bottom-up synthetic approach afforded by an inorganic NP scaffold at its core [4-6]. In this computation-guided design, biocompatible polyethylene glycol (PEG) polymers synthesized with one end of the PEG chain chemically linked to a lipid molecule (attached at the lipid head group) are covalently bound to the surface of an inorganic NP core. The inorganic core could be made of gold, silica, superparamagnetic iron oxide, and many other nanomaterials with high monodispersity and different shapes. The addition of free lipids to PEGylated particles induces the automatic absorption and nucleation of a lipid bilayer, due to the hydrophobic nature of lipid tails. In this way, a core-shell structure is built out from the inorganic NP, where the PEG polymer is sandwiched between the core and lipid bilayer shell. The surface chemistry of this core-PEG-lipid shell (CPLS) NP is identical to corresponding liposomes, which provides a logical route to their biocompatibility and makes them a natural fit as drug delivery vehicles. In direct comparison with liposomes, our large-scale molecular simulations and theoretical analysis confirm that the proposed CPLS NPs have the following advantages [1-5]: (1) size uniformity due to constraints applied through the brush height of tethered PEG polymers at a NP core, resulting in more uniform size distributions of CPLS NPs and therefore better control over total loading of cargo molecules and improved batch to batch reproducibility; (2) the ability to scaffold non-spherical shapes will be more easily achieved as the shape of inorganic core is not limited to a sphere and can thus be templated on a rod, disk or a multi-faceted shape, an important assembly feature as high aspect ratio materials have been shown to significantly bias the route of entry into cells [4]; (3) enhanced stability during administration, particularly when administered through vasculature under high shear flow, as the CPLS’s liposomal-like structure will have added stability afforded by its solid inorganic NP core and covalent linkage (via PEG-lipid anchor) to the NP surface [5] and (4) the potential for multifunctional properties and development as a ‘theranostic’ material, where the lipid vesicle can account for drug carrier capabilities and the inorganic core can be useful as a diagnostic imaging agent, either as an optical tag or core for photothermal therapy. The research impact of this work will be the forging of boundaries between multidisciplinary researchers for enriching an area of research that is ‘cyber-enabled nanomedicine design’ through the convergence of molecular simulation, molecular mean-field theory, and complementary experiments (conducted by Dr. Jessica Rouge’s group at UConn). The innovations obtained from this research have the potential to redesign the way we build nanoscale liposomal materials from the bottom up, offering improved capabilities that embody the desirable and necessary properties for effective drug delivery and diagnostic applications.

[1] Shen, Zhiqiang, David T. Loe, Joseph K. Awino, Martin Kröger, Jessica L. Rouge, and Ying Li. “Self-assembly of core-polyethylene glycol-lipid shell (CPLS) nanoparticles and their potential as drug delivery vehicles.” Nanoscale 8, no. 31 (2016): 14821-14835.

[2] Shen, Zhiqiang, Huilin Ye, Martin Kröger, and Ying Li. “Self-assembled core–polyethylene glycol–lipid shell nanoparticles demonstrate high stability in shear flow.” Physical Chemistry Chemical Physics 19, no. 20 (2017): 13294-13306.

[3] Shen, Zhiqiang, David T. Loe, Alessandro Fisher, Martin Kröger, Jessica L. Rouge, and Ying Li. “Polymer stiffness governs template mediated self-assembly of liposome-like nanoparticles: simulation, theory and experiment.” Nanoscale 11, no. 42 (2019): 20179-20193.

[4] Shen, Zhiqiang, Huilin Ye, Xin Yi, and Ying Li. “Membrane wrapping efficiency of elastic nanoparticles during endocytosis: Size and shape matter.” ACS nano 13, no. 1 (2018): 215-228.

[5] Shen, Zhiqiang, William Baker, Huilin Ye, and Ying Li. “pH-Dependent aggregation and pH-independent cell membrane adhesion of monolayer-protected mixed charged gold nanoparticles.” Nanoscale 11, no. 15 (2019): 7371-7385.

Area III: Predicting Nanoparticle Transport in Human Vasculature through High-Performance Computing

      The encapsulation of drug molecules and contrast agents into nanoparticles (NPs) can provide significant improvements in pharmacokinetics, toxicity, and biodistribution compared to freely administered drug molecules. The therapeutic efficacy of NP-based drug carriers is determined by the proper concentration of drug molecules at the lesion site. NPs need to be delivered directly to the diseased tissues while minimizing their deposition/uptake by other tissues, thereby reducing the potential harm to healthy tissue. From the clinical point of view, it is important to predict: 1) The distribution and concentration of injected NPs along the vascular pathway (evaluate damage to healthy tissues) and in targeted region (evaluate delivery efficacy); 2) An optimal design, dose, and concentration of NPs to achieve the dual goals of minimal damage to healthy tissues and maximal curing dose for the targeted region. The NP-mediated drug delivery in vascular system involves the interplay of transport, hydrodynamics, and multivalent interactions with targeted endothelial wall. It is very challenging to explore these phenomena experimentally in vivo, due to the small size of NPs, the dynamic drug delivery process, and complex vascular environment. Therefore, computational modeling of this process for NPs under different vascular flow conditions is crucial to elucidate the effects of NP property and dosage on their circulation, distribution, and clearance in vivo.

To address this issue, we have developed a hybrid Lattice Boltzmann method (LBM) and molecular dynamics (MD) computational approach for modeling NP transport in human vasculature [1-5]. Such an efficient computational method can be applied to examine the transport, interaction, and deposition of NPs within a vascular network by coupling the LBM and MD simulations. The realistic geometry of the vascular network and fluid dynamics of blood flow have been accurately captured through the LBM. While the microscopic interactions between NPs and red blood cells (RBCs) within blood flow and adhesion of NPs to vessel wall have been resolved through the MD simulation. A robust and efficient coupling interface is crafted to hybrid LBM and MD solvers together. Specifically, we have explored how the size, shape, surface property, and stiffness (4‘S’) of NPs can influence their vascular transport behaviors. For instance, we find that nanoworms with length of 8 μm and moderate stiffness are the optimal choice as drug carriers for circulating within a normal vascular network due to their lower near-wall margination, in comparison to rigid spherical particles [3]. According to our simulation results, tuning the length and stiffness of nanoworms is the key to design drug carries with reduced near-wall margination within normal vascular networks and extend their blood circulation time. Our work presents a number of unique features, both at the level of high-performance computing technology and in terms of physical/computational modeling for science and engineering applications in nanomedicine. The Texas Advanced Computing Center (TACC) has recently reported our study in the article, “targeting-tumors-with-nanoworms”.

[1] Ye, Huilin, Zhiqiang Shen, and Ying Li. “Computational modeling of magnetic particle margination within blood flow through LAMMPS.” Computational Mechanics 62 (2018): 457-476.

[2] Ye, Huilin, Zhiqiang Shen, and Ying Li. “Interplay of deformability and adhesion on localization of elastic micro-particles in blood flow.” Journal of Fluid Mechanics 861 (2019): 55-87.

[3] Ye, Huilin, Zhiqiang Shen, Le Yu, Mei Wei, and Ying Li. “Anomalous vascular dynamics of nanoworms within blood flow.” ACS biomaterials science & engineering 4, no. 1 (2018): 66-77.

[4] Ye, Huilin, Zhiqiang Shen, Mei Wei, and Ying Li. “Red blood cell hitchhiking enhances the accumulation of nano-and micro-particles in the constriction of a stenosed microvessel.” Soft Matter 17, no. 1 (2021): 40-56.

[5] Ye, Huilin, Zhiqiang Shen, Weikang Xian, Teng Zhang, Shan Tang, and Ying Li. “OpenFSI: A highly efficient and portable fluid–structure simulation package based on immersed-boundary method.” Computer Physics Communications 256 (2020): 107463.