(2aq) Accurate Computational Design of Programmable 3D Protein Crystals and Capsids

My research centers on programming self-assembly of nanoscale architectures through supramolecular interactions, applying molecular engineered oligonucleotides and computational protein designs.

My PhD work explored two fundamental aspects of colloidal crystal engineering of nanoparticles with programmable DNA- based interactions: 1) I developed a new class of phosphate-based ligands (e.g. lipids, DNA) that selectively modify the external surface of porous metal-organic framework (MOF) nanoparticles through coordination bonds to open metal nodes. I further explored their applications in intracellular protein-delivery and colloidal crystal engineering. 2) I discovered particle analogs of electrons in colloidal crystals. Our experiments and MD simulation demonstrate that highly dynamic DNA bonds enable tiny particles to roam freely across unit cells within a superlattice of complementary large particles, thereby adopting highly delocalized spatial distribution. This "particle-bond" duality challenges the prevailing consensus regarding colloidal crystals, which presumes that particle building blocks remain stationary and are situated in clearly defined lattice positions. Furthermore, the concept of colloidal valency emerged from these studies. It is defined by the local coordination environment and diffusivity of these small particles, leading to a set of design principles and discovery of nine distinct metallic and intermetallic phases, including an unprecedented gyroid nanoparticle superlattice.

During my postdoctoral research in the Baker lab. My objective is to create fully genetically encodable nanomaterials using computationally designed protein-protein interfaces. My first research project resolved the longstanding question of whether 3D protein crystals can be designed computationally. We predicted three self-assembled 3D protein crystals in silico and experimentally verified them, showing near-angstrom-level accuracy with their design models. In my second project, I developed a top-down design approach to construct self-assembled protein architectures using reinforcement learning (RL). This study was inspired by the RL algorithms used by AlphaGo, which excels in board games, such as Go. We transformed protein design tasks into games or puzzles with custom rules to be solved by AI. I expect that this innovative strategy will introduce a new framework for advancing molecular engineering in complex systems. Below please find the detailed abstract for these two studies:

Future project summary

• Designed crystalline proteinaceous subcellular factories

Structural properties of subcellular organelles regulate essential biological processes, including gene regulation, metabolism, and mechanotransduction, through precise temporal and spatial control over energy transfer and chemical transformation. Established upon my postdoc work on designed 2D and 3D protein crystals, I propose to further develop de novo synthetic subcellular organelles made of crystalline proteinaceous frameworks. These fully genetically encodable architectures are first computationally designed and tested ex situ, and then introduced to living systems, where additional spatial and temporal control can be achieved through rational design of gene circuits. My preliminary results already show that these de novo designed in cellulo crystals are highly biocompatible, form robustly in different cell lines, and do not interfere with regular cell divisions. In this vein, I plan to continue exploring two main directions: a) design functional in cell 3D protein crystals with enzymatic, energy harvesting, and inorganic nucleation sites; and b) couple inducible crystallization and disassembly with cell signaling events such as kinases and change in environmental pH and redox potential. To study these systems, I plan to take a decoupled approach to separate structural modules and functional motifs for increased tailorability and generality. I will computationally design 2-component crystallization systems that assemble through well-defined heterodimeric protein-protein interfaces. A heterodimer pair will be splitted and positioned adjacent to functional motifs separately, and then symmetrically arranged to form a lattice with gaps, which will be filled with de novo protein backbones through generative protein design models, such as RFdiffusion. Such a modular bond-centric approach enables straightforward control over protein crystallization, with tunable functional site spacing, channel dimensions, and heterodimer binding strength. As a proof-of-concept, de novo designed protein crystals will be designed and tested as host materials for in situ nucleation and encapsulation of light sensitive nanoparticles inside cells, such as CdSe and InP quantum dots. The in situ synthesis of inorganic nanoparticles inside protein crystals are promoted by deliberate installation of peptides or designed proteins that serve as nucleation sites. These bioinorganic hybrid platforms shall selectively impact redox-dependent metabolic pathways (e.g. shikimic acid) through photoexcited electron generation from inorganic nanoparticles, and more broadly provide solutions to a range of photosynthesis and biomanufacturing related problems for sustainability. Further, e.coli and yeasts expression will first be tested as model living systems, because their genetic modifications are well-established with readily available protocols. Finally I will move to more complex bio-production systems, such as filamentous fungi with high protein expression level and mammalian cells for post-translational modifications.

• Modular virus-like nanoparticle for delivery

The shell forming proteins of microcompartment in bacteria and viral capsids are highly evolved self-assembled architectures that specialize in isolating enzymatic reactions, protecting and delivering its genetic materials for delivery purposes. State-of-the art research efforts have primarily focused on engineering and repurposing viral-vectors for delivery purposes, including much studied AAV and lentivirus. Despite their successful applications as vaccines and delivery vehicles for genetic materials, they are limited by their loading capacity, immune response, and challenges during manufacturing. My postdoc work has shown that mini-capsids can be designed in a top-down manner with RL backbone sampling, where particle porosity can be controlled by imposing geometric constraints. However, it is still challenging to design protein nanoparticles larger than 40 nm, and with functional motifs for packaging and endosomal escape. In this vein, I propose to launch a modular design effort to computationally generate robust self-assembled large de novo virus-like nanoparticles (50 ~100 nm) as promising delivery vehicles. I plan to tackle this problem with three steps: first, develop algorithms and a computational pipeline to generate capsid backbones that integrate all functional motifs (targeting, encapsulation, endosomal escape, etc.) into two to three feasible building block partners. In particular, I’m interested in using generative protein design models in combination with RL guided constrained structural sampling, which ensures low porosity on particle shells. Another existing challenge is that capsid formation often relies on self-assembly cooperativity, where interfaces are weak, loopy, and highly polar, distinguishing them from conventional protein-protein oligomeric interfaces. To address this, I plan to fine-tune the ProteinMPNN model to learn natural capsid sequence preferences, for improved interface sequence design targeting relatively weak interfaces. Second, experimentally expression and screening for correctly assembled particles shall be performed using a high-throughput approach tagged by DNA barcodes, followed by pull-down assay on size exclusion based, and further analyzed using mass spectrometry and sequencing technologies. Promising candidates will be characterized biochemically for functions, and structurally with cryo-electron microscopy to verify design success. Third, I plan to take advantage of well-established functional assays to evaluate and optimize a variety of potential applications of these de novo VLPs, including nucleic acids and protein encapsulation, cytotoxicity, cell entry, and cargo release mechanism. Finally, if positive outcomes could be realized by above mentioned in vitro experiments, I will move on to study and optimize cargo delivery and immunogenicity of these constructs in animal models for potential gene editing and protein therapeutics.

Teaching Interests

In my understanding, the hallmarks of an exemplary educator encompass the dual roles of being a proficient communicator and nurturing mentor. As a past college student, I reaped significant benefits from instructors who articulated complex concepts with discernable logic and unambiguous clarity. In my later role as a teaching assistant in introductory and advanced organic chemistry labs and as a mentor to more than 10 undergraduate and graduate researchers, I have formed the belief that efficient instruction is a holistic reflection of a profound comprehension of the subject matter and an innate capacity for empathetic perception. The former guarantees the quality of educational content, while the latter enhances the potency of knowledge dissemination. Consequently, a potent educator should embody key traits akin to those of a proactive mentor, investing time and effort into understanding their students' unique profiles, discerning their needs, and thus identifying the optimal method to spur their active participation in the learning process. I firmly believe that fostering a comfortable and accepting classroom environment is paramount, where diversity forms an integral component of successful science education and research.

Looking ahead, I am committed to dedicating my efforts towards educating future generations of chemical engineers. The academic background I have gained through my graduate studies and postdoc training in chemistry, nanotechnology, and computational protein design equips me with a solid foundation in experimental physical sciences and state-of-the art computational methodologies. I am confident to provide instruction across a broad spectrum of undergraduate-level courses, such as organic, inorganic, and polymer chemistry, biochemical engineering, thermodynamics, and more. If granted the opportunity, I aim to develop a course focused on the principles of chemical process safety in manufacturing, as well as rules and work ethics in the modern, digital era. For graduate-level courses, I intend to incorporate my expertise and offer core courses like protein engineering, machine learning and data science in chemical engineering, and elective courses emphasizing colloids, computational protein design, or synthetic methodologies in nanomaterials.