Trung D. Nguyen

Home Research Software Notes
RESEARCH
Cover images See my publications on Google Scholar or ORCID.

VISION

Future materials and devices will adapt themselves in response to the environment to perform their functions without compromising their efficiency. Furthermore, they will self-repair and self-improve by "learning" from defects and failures during their service lifetime. Even better, they can augment living cells, and replace damaged ones whenever necessary. In other words, I am interested in finding rules to create materials and devices that mimick living matter in the operational sense. Imagine such materals can be used for applications ranging from smart medical devices, cellular therapies for treating human neurodegenerative disorders, to safer batteries with high energy density. It is evident that many fascinating opportunites for engineering such materials and devices can be found at the atomic and molecular levels; and with them fundamental challenges are plenty.

Towards this vision, I have shown in my work many fascinating possibilities:

Essentially, my studies address the following fundamental questions:

1. Which nanostructures could be promising candidates for materials with a given set of properties?

Identifying functional nanostructures is certainly challenging. Throughout human history, we have witnessed how Nature has evolved to select structures that are more energy efficient, that are moving faster, and that are adaptive to environmental changes. My research is devoted to building upon such ideas for designing nanomaterials that mimic biological structures, first in terms of structural features, then in terms of their ability to reconfigure in response to changes in the environment.

✔ Percolated networks formed by random ionomers

✔ Uniform-sized clusters by like-charged nanoparticles and proteins

✔ Helical and tubular structures from polymer tethered nanorods

2. How can we engineer and control nanoscale interactions for targeted nanostructures?

At nanometer scale, it is well known that these building blocks interact via the van der Waals dispersion and electrostatic forces, and/or effective hydrophobic/hydrophilic interactions. While these interactions are non specific in nature, it is hard to predict how they are realized in specific systems and how they will dictate the collective behavior of the building blocks. For multicomponent systems, it becomes even more challenging because of the non-uniformity in the chemical and physical properties across different length scales.

To address these challenges, we have employed and developed all-atom, united atom and coarse-grained models and performed computer simulations to reveal a lot of fascinating insights into phenomena observed in experiment and predict the structural and dynamic behaviors of the resulting assemblies from these nano building blocks. The ultimate goal here is to find design rules for the interactions between the building blocks for specific structures.

✔ Patterning hydrophobic polymer segments for encapsulating proteins in unfavorable solvents for protein stabilization

✔ Tuning the electrostatic and excluded volume interactions between colloids and nanoparticles for biosensing

✔ Combinging the nanoscale interactions with spatial confinement at solid-liquid interfaces for 3D printing and coating

3. How can we make the assembled nanostructures respond to environmental cues like biological matter?

This is the key challenge towards engineering biomimetic materials, those that can respond to changes in the environment such as pH, temperature, ionic strength or light in prescribed manners (look at the skin color of betta fish for instance). Traditionally, approaches to engineering materials from bottom up involve designing static nanoassemblies in the sense that the assembled structures are, and should be, thermodynamically stable states.

Here we relax that requirements by investigating strategies for making nano-assemblies that transform spontaneously between different morphologies in response to environment cues. We show that changing dielectric constrast, making protein-mimicking polymers, varying the effective interactions between parts of the building blocks or morphing their shapes open a host of new possiblities for reconfigurable nanostructures, which in many ways resemble biological matter. Indeed, the ability to reconfigure or adapt the environmental changes is one of the prerequisites for the materials and devices to self-repair and self-improve.

✔ Tuning solvent ionic strength and polarity

✔ Shifting particle shapes

To tackle the above questions using modeling and computational tools, I have helped develop multiscale models and new algorithms that capture the essential physics of the interested systems, and implemented them in high-performance computing software packages.

Method development

My research incorporates multiscale modeling and computational methods to take advantage of high-performance computing, advanced sampling and machine learning techniques. In the data science era, while most research in the field are concerned with how to apply machine learning techniques to the design problems in specific domains (like for polymers, colloids, drug molecules), the fundamental problem lie in how to generate high-quality datasets. For molecular and nanoscopic systems, the challenge to obtain high-quality datasets originate from the probabilistic nature of the output given a set of input parameters. For instance, the microscopic configurations observed in a however long Molecular Dynamics simulation could simply be limited to a narrow region of the phase space that is available to the simulated system, and the probability of observing other, likely more relevant, configurations, is vanishingly low.

Although techniques to improve the statistical quality of the output have been proposed for years, incorporating them into machine-learning workflows is computationally expensive, so most proof-of-concept studies to date either rely on existing datasets (if any), or resort to "lazy" computations (meaning poor statistics).

To address such computational challenges with machine learning enabled design, I have purused the following research directions:

  1. leveraged the power of hybrid CPU/GPU architectures to accelerate the speed of Molecular Dynamics simulations,
  2. implemented and developed various algorithms to improve sampling of microscopic systems,
  3. coupled particle-based with continuum calculations to bridge different time scales, and
  4. employed pattern recognition and combined different supervised classifiers for structural identification problems in molecular systems.

The representative studies for this theme of research include: