Decoding of in vivo two-photon imaging data in mouse motor cortex

如果事先有准备，有目的的动作会更有效。这一运动准备过程被认为是由运动皮层介导的，至少部分是由运动皮层介导的。支持这一观点的是，运动皮层中的神经元在动物准备和执行运动时调节放电率。然而，我们不知道运动皮层中的神经元回路是如何将准备转化为行动的。在目前的建议中，我们将使用体内双光子钙成像研究准备和执行前臂运动的小鼠运动皮层的神经活动。这种方法将使我们能够对运动皮层中神经回路的活动进行成像。为了从丰富的成像数据中提取更多的信息，我们将使用基于机器学习的线性解码器。首先，我们将对成像数据进行解码，以预测反应时间和其他运动相关参数。通过这个过程，解码器通过确定每个神经元的权重因子来识别编码相关信息的神经元。通过检查在准备和执行期间不同时间点的权重因子，我们将揭示运动准备和运动命令的神经表征的时间动态。其次，我们的目标是揭示回路的活动状态如何一刻一刻地变化，直到执行前臂运动。为了揭示这种活动状态的动态，我们将使用输入变量作为特定帧的成像数据来训练解码器，并在下一帧中预测目标神经元的活动。通过确定每个神经元的权重因子，我们将确定有助于预测目标神经元活动的神经元。通过这些项目，我们将以定量的方式将神经回路活动的动态变化与动物的行为联系起来，这将有助于更好地理解人类行为背后的计算过程。

Purposeful movements are produced more effectively if preceded by preparation. This process of motor preparation is considered to be mediated, at least partially, by the motor cortex. Supporting this view is that the neurons in the motor cortex modulate discharge rate as animals prepare and execute movements. However, we do not know how circuits of neurons in the motor cortex operate to convert preparation into action. In the current proposal, we will investigate the neural activity of motor cortex in mice that prepare and execute forearm movements using in vivo two-photon calcium imaging. This method will allow us to image the activity of neural circuits in the motor cortex. To extract much information out of the rich imaging data, we will use machine learning based linear decoder. First, we will decode the imaging data to predict the reaction time and other movement related parameters. Through this process, the decoder identifies the neurons that encode the relevant information by determining the weight factors for each neuron. By examining the weight factors at different time-points during the preparatory and execution period, we will reveal the temporal dynamics of the neural representation of the motor preparation and movement commands. Second, we aim to uncover how the activity states of the circuits change moment by moment until the execution of the forearm movements. To reveal this dynamics of the activity state, we will train the decoder with the input variables as the imaging data of a particular frame, and predict the activity of a target neuron in the following frame. By determining the weight factors for each neuron, we will identify the neurons that contribute in predicting the activity of the target neuron. Through these projects, we will link the dynamic change of the neural circuit activity and the animal¿s behavior in a quantitative fashion, which will lead to the better understanding of the computational processes that underlie our actions.