Data-driven Multiscale Modeling of Signal Processing in Bacterial Chemotaxis

变化点分析是一种不需要噪声模型知识的分析方法，用于检测时间序列中可观测值和线性趋势变化的时刻。模型自由变化点检测已应用于各种单分子时间序列，如单蛋白马达旋转测量、单离子通道膜片钳测量中的门控电流波动等，以捕获传统阈值和分形方法无法解决的事件开关。变化点分析的结果是系统在不同状态下的停留时间统计，如旋转蛋白质马达的暂停和旋转状态，离子通道门控的亚电导状态，鞭毛马达的顺时针和逆时针旋转模式，细菌趋化性的不同扩散模式等。作为项目的下一步，所得到的停留时间统计数据将作为输入来推断系统的潜在动力学方案，以揭示隐藏的动力学状态和状态之间的转换。另一方面，将用于量化和检测多元时间序列之间信息流的“传递熵”的数学形式推广到检测“时间相关”流。在将新的数学形式应用于真实的细菌趋化性时间序列之前，该理论目前使用从简单模拟模型生成的时间序列来测试其性能和有效性。

Change point analysis that does not require the knowledge of noise model has been developed to detect the time instants at which the observable value and the linear trend change in a time series. The model free change point detection has been applied to various single molecule time series, such as from single protein motor rotary measurements, gating current fluctuation in single ion channel patch clamp measurements, etc., to capture event switching that cannot be resolved using conventional thresholding and binning methods. The outcome from the change point analysis is the dwell time statistics of the system residing in various states, such as the pause and rotation states in rotary protein motor, sub-conductance states in ion channel gating, clockwise and counter-clockwise rotation modes in flagellar motor, different diffusive modes in bacterial chemotaxis, etc. As a next step of the project, the resulting dwell time statistics will then serve as the input to infer the underlying kinetic scheme of the system to reveal hidden kinetic states and transitions among the states.On the other hand, the mathematical formalism of "transfer entropy" to quantify and detect information flow between multivariate time series has been generalized to detect "time dependent" flows. Before applying the new mathematical formalism to real bacterial chemotaxis time series, the theory is currently tested using time series generated from simple simulation models to benchmark its performance and validity.