One of the keys for the existence of life is metabolism. The complex collection of metabolic reactions enables constant synthesis and degradation of the building blocks of life, dynamically remolds and regulates the life’s molecular organizations, and keeps life staying in a non-equilibrium state. The monitoring of the kinetics of the analog metabolic reactions is critical to profile the life-like dynamic behaviors and characteristics of the complex chemical systems. This is also a complicated, onerous, and time-consuming process, affected by many factors, e.g., temperature, concentration, pH, reaction time, etc. The unexpected, abnormal fluctuations of various factors may generate data with low quality which may result in a failure in constructing kinetic models. In settings faced with the abnormal fluctuations of factors, detection of such fluctuations holds the potential to increase the robustness in building kinetic models of the complex chemical systems. The kinetics can be modeled as streaming, multi-dimensional time series, where each factor can be regarded as a dimension in time series. Thus, we can model the detection of the abnormal fluctuations of factors as a correlated time series outlier detection problem and design a dynamic outlier detection and interpretation framework, which can dynamically identify suitable factors reflecting the abnormal fluctuations of the kinetics of the metabolic reaction cycles and the dynamic behaviors of the complex chemical systems.
