Chun-Biu Li: Learning the (Nano)-Machines with Machine Learning from an Information-theoretic Approach

Abstract:
Discrete finite-level time series are often observed in biophysics experiments to extract the dynamical and kinetic information on the working mechanisms of a single (or multiple) bio- and nano-molecular machine(s), such as in ion-channel gating kinetics measurements, single enzymatic turnover experiments, observations of intermittent blinking of quantum dots, imaging of stepwise movements of single supermolecular motor in living systems, and assessments of association/dissociation kinetics in cell signaling processes, etc. Statistics of the dwell-time time series, the stationary state distributions associated with the chronological sequence of the lengths of time that the system dwells at each level, have been studied to infer the underlying dynamics and kinetics of the system.

However, it is well known that the underlying kinetic scheme, a hidden Markov network composed of states and state transitions, cannot be identified uniquely from the observed dwell-time statistics. This is because some states are aggregated and the associated transitions of the underlying kinetic scheme are hidden which cannot be resolved by finite-level measurements. In this talk, I will present an information theoretic framework to quantify the amount of excessive (extra) information contained in a given kinetic scheme that is not warranted by the observed dwell-time statistics. The kinetic scheme capturing all information of the observed dwell-time statistics with minimum excessive information can be uniquely identified and constructed, and it is regarded as the most objective kinetic network one can extract from the data. The minimum excessive information enables us to scrutinize which observable provides more information to construct a kinetic scheme closer to the underlying true scheme. The method is applied to a single molecule enzymatic turnover experiment, and the origin of dynamic disorder is discussed in terms of the network properties of the constructed kinetic scheme.