Kaiwen Sheng

All Publications

• Remodelling of corticostriatal axonal boutons during motor learning. Nature Sheng, M., Lu, D., Roth, R. H., Hwang, F. J., Sheng, K., Ding, J. B. 2025

  Abstract

  Motor skill learning induces long-lasting synaptic plasticity at dendritic spines1-4 and at the outputs of motor cortical neurons to the striatum5,6. However, little is known about corticostriatal axon activity and structural plasticity during learning in the adult brain. Here, using longitudinal in vivo two-photon imaging, we tracked thousands of corticostriatal axonal boutons in the dorsolateral striatum of awake mice. We found that learning a new motor skill dynamically regulated these boutons. The activities of motor corticostriatal axonal boutons exhibited selectivity for rewarded movements (RM) and unrewarded movements (UM). Notably, boutons on the same axonal branches showed diverse responses during behaviour. Motor learning significantly increased the proportion of RM boutons and reduced the heterogeneity of bouton activities. Moreover, motor learning induced profound structural dynamism in boutons. By combining structural and functional imaging, we saw that newly formed axonal boutons were more likely to exhibit selectivity for RM and were stabilized during motor learning, whereas UM boutons were selectively eliminated. These findings reveal a novel form of plasticity in corticostriatal axons and show that motor learning drives dynamic bouton reorganization to support motor skill acquisition and execution.

  View details for DOI 10.1038/s41586-025-09336-w

  View details for PubMedID 40739352

  View details for PubMedCentralID 2844762

• System-level time computation and representation in the suprachiasmatic nucleus revealed by large-scale calcium imaging and machine learning. Cell research Wang, Z., Yu, J., Zhai, M., Wang, Z., Sheng, K., Zhu, Y., Wang, T., Liu, M., Wang, L., Yan, M., Zhang, J., Xu, Y., Wang, X., Ma, L., Hu, W., Cheng, H. 2024

  Abstract

  The suprachiasmatic nucleus (SCN) is the mammalian central circadian pacemaker with heterogeneous neurons acting in concert while each neuron harbors a self-sustained molecular clockwork. Nevertheless, how system-level SCN signals encode time of the day remains enigmatic. Here we show that population-level Ca2+ signals predict hourly time, via a group decision-making mechanism coupled with a spatially modular time feature representation in the SCN. Specifically, we developed a high-speed dual-view two-photon microscope for volumetric Ca2+ imaging of up to 9000 GABAergic neurons in adult SCN slices, and leveraged machine learning methods to capture emergent properties from multiscale Ca2+ signals as a whole. We achieved hourly time prediction by polling random cohorts of SCN neurons, reaching 99.0% accuracy at a cohort size of 900. Further, we revealed that functional neuron subtypes identified by contrastive learning tend to aggregate separately in the SCN space, giving rise to bilaterally symmetrical ripple-like modular patterns. Individual modules represent distinctive time features, such that a module-specifically learned time predictor can also accurately decode hourly time from random polling of the same module. These findings open a new paradigm in deciphering the design principle of the biological clock at the system level.

  View details for DOI 10.1038/s41422-024-00956-x

  View details for PubMedID 38605178

  View details for PubMedCentralID 10717972

• U-RISC: An Annotated Ultra-High-Resolution Electron Microscopy Dataset Challenging the Existing Deep Learning Algorithms FRONTIERS IN COMPUTATIONAL NEUROSCIENCE Shi, R., Wang, W., Li, Z., He, L., Sheng, K., Ma, L., Du, K., Jiang, T., Huang, T. 2022; 16: 842760

  Abstract

  Connectomics is a developing field aiming at reconstructing the connection of the neural system at the nanometer scale. Computer vision technology, especially deep learning methods used in image processing, has promoted connectomic data analysis to a new era. However, the performance of the state-of-the-art (SOTA) methods still falls behind the demand of scientific research. Inspired by the success of ImageNet, we present an annotated ultra-high resolution image segmentation dataset for cell membrane (U-RISC), which is the largest cell membrane-annotated electron microscopy (EM) dataset with a resolution of 2.18 nm/pixel. Multiple iterative annotations ensured the quality of the dataset. Through an open competition, we reveal that the performance of current deep learning methods still has a considerable gap from the human level, different from ISBI 2012, on which the performance of deep learning is closer to the human level. To explore the causes of this discrepancy, we analyze the neural networks with a visualization method, which is an attribution analysis. We find that the U-RISC requires a larger area around a pixel to predict whether the pixel belongs to the cell membrane or not. Finally, we integrate the currently available methods to provide a new benchmark (0.67, 10% higher than the leader of the competition, 0.61) for cell membrane segmentation on the U-RISC and propose some suggestions in developing deep learning algorithms. The U-RISC dataset and the deep learning codes used in this study are publicly available.

  View details for DOI 10.3389/fncom.2022.842760

  View details for Web of Science ID 000791580200001

  View details for PubMedID 35480847

  View details for PubMedCentralID PMC9038176

• Siamese Network-Based All-Purpose-Tracker, a Model-Free Deep Learning Tool for Animal Behavioral Tracking FRONTIERS IN BEHAVIORAL NEUROSCIENCE Su, L., Wang, W., Sheng, K., Liu, X., Du, K., Tian, Y., Ma, L. 2022; 16: 759943

  Abstract

  Accurate tracking is the basis of behavioral analysis, an important research method in neuroscience and many other fields. However, the currently available tracking methods have limitations. Traditional computer vision methods have problems in complex environments, and deep learning methods are hard to be applied universally due to the requirement of laborious annotations. To address the trade-off between accuracy and universality, we developed an easy-to-use tracking tool, Siamese Network-based All-Purpose Tracker (SNAP-Tracker), a model-free tracking software built on the Siamese network. The pretrained Siamese network offers SNAP-Tracker a remarkable feature extraction ability to keep tracking accuracy, and the model-free design makes it usable directly before laborious annotations and network refinement. SNAP-Tracker provides a "tracking with detection" mode to track longer videos with an additional detection module. We demonstrate the stability of SNAP-Tracker through different experimental conditions and different tracking tasks. In short, SNAP-Tracker provides a general solution to behavioral tracking without compromising accuracy. For the user's convenience, we have integrated the tool into a tidy graphic user interface and opened the source code for downloading and using (https://github.com/slh0302/SNAP).

  View details for DOI 10.3389/fnbeh.2022.759943

  View details for Web of Science ID 000781044900001

  View details for PubMedID 35309679

  View details for PubMedCentralID PMC8931526

• FlexTensor: An Automatic Schedule Exploration and Optimization Framework for Tensor Computation on Heterogeneous System Zheng, S., Liang, Y., Wang, S., Chen, R., Sheng, K., ACM ASSOC COMPUTING MACHINERY. 2020: 859-873

  View details for DOI 10.1145/3373376.3378508

  View details for Web of Science ID 000541369300055