Research

Our research program aims to develop methods that address fundamental questions about how cell functions give rise to behavior and how their dysfunctions can lead to diseases. For example, we are working on building techniques to understand the collective workings of neurons in the brain that generate cognitive functions, to understand the cellular malfunctions that result in disease, and to study how neurons organize into the brain's complex structure. A significant challenge we face is the spatiotemporal dilemma: the human brain contains 85 billion neurons spread across several centimeters, with each neuron is only about 10-100 µm in size. Furthermore, while individual neurons operate on millisecond scales, brain activities such as emotion, cognition, and learning can span from days to years. Adding to this complexity, the brain comprises diverse neuron types, each with unique gene expressions, morphologies, and electrical behaviors.

To address these challenges, we are transforming high-performance electronics into soft and flexible bioelectronics that can be seamlessly integrated with biological tissues. We are combining bioelectronics with advanced genetic and genomic engineering, and computational tools to create a multimodal, scalable bioelectronics for tissue interfaces. Our research include interfacing with numerous cells at a single-cell resolution across 3D tissue volumes, tracking and modulating activities from the same cells at high resolution in a long-term stable manner, and targeting specific cell types genetically. The ultimate goal is to develop a platform capable of interfacing with the entire 3D brain over extended periods. We've achieved significant progress in four technological areas:

1. Tissue-like electronics: Our team has pioneered flexible and soft materials compatible with advanced semiconductor chip fabrication processes to fabricate scalable flexible bioelectronics (Nature Nano, 2015, Nature Nano., 2023). With advanced bio-integration techniques, we can seamlessly merge these electronics with 3D tissues both in vitro and in vivo. These electronics avoid chronic tissue damage, immune responses, and drifting, allowing long-term cell activity recording from the same neurons (Nature Neurosci. 2023). We have extended these techniques to design next-generation flexible brain-computer interfaces for various animal models and human patients (Nature Electron. 2023).

2. Cyborg engineering: To integrate electronics within 3D tissues, we've spearheaded a unique direction using the 2D-to-3D organogenesis of tissues during its development to implant flexible and stretchable mesh electronics within them. This method has been applied extensively in human stem-cell derived organoids, including brain, heart, and pancreatic types, as well as animal embryos. We've achieved long-term, stable electrical characterizations of heart (Nano Lett., 2019, Sci. Adv., 2023) and brain organoids (Adv. Mater., 2022) during their maturation. Our work with animal embryos has for the first time unveiled the evolutionary trajectory of brain activity during development. Significantly, we pioneered capturing single-cell action potentials in pancreatic organoids, offering insights on their functional maturation for diabetes treatment.

3. In situ electro-sequencing: To correlate cell gene expression with electrical recordings, we've integrated cutting-edge spatial transcriptomics (in situ single-cell RNA sequencing) (Science, 2018) into our electronic platform as “in situ electro-sequencing” technique platform (Cell, 2023). Our team has also developed machine-learning computational tools to analyze and integrate spatially resolved multimodal gene expression and electrical data at the single-cell level (Nat. Commun., 2021, Nat. Commun., 2023, Cell, 2023), jointly defining cell states and developmental trajectories; identifying gene-to-electrophysiology relationships throughout tissue development; and accurately reconstructing the evolution of gene expression profiles based on long-term stable electrical measurements. We are applying in situ electro-sequencing to create spatiotemporal multimodal maps in electrogenic tissues for discovery of cell types and gene programs responsible for cell function and dysfunction.

4. AI for long-term and multimodal data analysis. Our team has developed a multimodal deep learning-based computational pipeline to analyze long-term electrical data (BioRxiv, 2024). To process and analyze multimodal data from in situ electro-sequencing, we created automated machine learning pipelines that integrate multitask and multimodal frameworks. These pipelines enable the automated generation of cell-type-specific feature-to-feature correlations across modalities, including gene expression, morphology, and electrical phenotypes (Nat. Commun. 2023). Additionally, we extended this capability by leveraging gene expression and spatial data to align and integrate measurements from millions of cells. This approach allows for the reconstruction of 3D tissue or even organ-level multimodal structures, providing unprecedented insights into complex biological systems.

In addition, we've successfully combined these four innovation directions into a unified platform, setting the stage for groundbreaking advances in the following research directions:

1. Next generation brain-computer interfaces (BCIs). The brain is a dynamic system – neural activities continuously change during sensing, learning, memory, disease progression, and aging. Stable recordings from the same neurons are crucial to understanding and decoding these activities in both health and disease states. Traditional BCIs, however, employing rigid electronics for brain interfacing, leading to inconsistent long-term neural activity recordings. Our flexible electronic brain interfaces fundamentally address these challenges. Leveraging this platform, we aim to (1) Decipher intrinsic neural dynamics: understand the intrinsic dynamics of neural activities during development, learning, memory, and ageing, ensuring precise brain state decoding; (2) Build AI-controlled neuroprosthetics: utilize consistent recordings as training data for machine learning algorithms, paving the way for personalized, autonomous, and energy-efficient decoding in next-generation neuroprosthetics; (3) Develop AI-driven deep brain stimulators: leverage the gliosis-free electronics-tissue interface and stable recording signals to build next-generation deep brain stimulator with AI-driven closed-loop control to specifically modulate individual neuron and neural circuit behaviors, mitigate irregular neural activities in neurological disorders, guide neural progenitor cell movement, and modulate a large number of neurons based on their spatiotemporally orchestrated activity patterns. 

2. NeuroAI: Integrating Nature and Machine Intelligence. Understanding intrinsic, long-term neural dynamics in the brain is crucial for developing AI algorithms that possess the robustness and efficiency of biological systems. Traditional techniques, however, are unable to capture activities from the same cells over extended periods, limiting the ability to uncover these intrinsic long-term neural dynamics. Our research addresses this limitation by introducing flexible BCIs capable of chronically and stably tracking single-neuron activities from the same cells over long time. This breakthrough provides unparalleled insights into the intrinsic dynamics of neural networks. By leveraging this knowledge, we are integrating these neural dynamics into the design of robust, efficient, and adaptive AI-based learning systems, creating a bridge between natural and artificial intelligence.

3. Bioelectronics for regenerative medicine and cell therapy. Cardiovascular diseases, neurological disorders, and diabetes collectively contribute to over 32% of all deaths in the past decade. The current medical treatments often fall short in tailoring treatments to individual patient needs, resulting in therapies that may be only partially effective and are accompanied by potential side effects. Stem cell-derived organoid technology has emerged as a promising alternative, offering patient-specific microtissues for personalized drug screening and cell sources for stem cell therapy. However, the assessment and control of organoid development, maturation, and functional specialization remain limited. Our research uniquely addresses these questions through the following innovations: (1) We have developed multimodal and multifunctional “tissue-like” bioelectronics capable of being integrated with organoids during development for chronically stable tracking and stimulation of organoid-wide multimodal (e.g., electrical, mechanical, and chemical) signals; (2) We have introduced “in situ electro-sequencing” that integrate spatial transcriptomics with electrical recording at single-cell level and created multitask and multimodal machine learning-based automated computational pipelines to process and analyze these complex datasets for drug screening, and (3) We are pioneering reinforcement learning-enabled closed-loop control to real-time, bidirectionally, and long-term stably control cellular activities across the 3D tissue to accelerate their functional maturation and specialization in both in vitro and in vivo setting.

4. Functional cell atlas. The spatial architectures of tissues are intrinsically linked to their function, which spans from the detailed shapes of cells to the diversity of cell types. The function of these cells is determined by a combination of factors, including cell gene expression, morphology, and connectivity. To fully comprehend tissue function, development, and disease, it is essential to correlate spatially resolved cell gene expression with morphological, physiological, and behavioral data. For example, one of the core missions of the US BRAIN initiative, akin to the Apollo project of the 1960s, is to create a brain atlas that spatially map all the cell gene expression, morphology, and electrical function in brains across different species. Our team has recently published a pioneering study (Nature, 2023): the first comprehensive brain-wide, single-cell gene expression map of the mouse brain. Building upon this, we have further developed a revolutionary AI tool that can seamlessly integrate our brain mapping data with other datasets, creating a comprehensive whole mouse brain cell atlas. This development is historic, offering us, for the first time, a detailed single-cell resolution gene expression map of the entire mouse brain. It also presents numerous opportunities to merge various brain modality mappings, especially these long-term, behavior-dependent electrical mapping from our lab into a functional brain cell atlas. Our current efforts are now focused on constructing this atlas, which promises to be an important resource in understanding brain functionality. We are also extending this functional atlas across different types of organs.