Machine Learning Models for Biological Processes
Biological processes such as cell cycle, organism development, and individual health are often controlled by a large number of molecules or biological components that interact and exchange information in a context-dependent manner. Over the course of these biological processes, there may exist multiple underlying "themes" that determine each molecule's function and relationship with other molecules. Such themes are both dynamic and stochastic. As a result, the molecular networks at each time point can undergo systematic rewiring and exhibit rich temporal structures instead of being invariant over time, as assumed in most current biological network studies.
In this talk, I will introduce multi-dimensional time-series models and point processes for such dynamic processes over networked biological systems. The two classes of models are complementary in the sense that times series models are suitable for characterizing biological processes over regular time-intervals, while point processes are more advantageous in modeling extreme activities of the involved biological entities. I will also show how one can estimate the parameters in these models from data, and uncover and interpret the biological networks behind these biological processes.