Learning Biological Regulatory Networks from Time Series with LFIT and Application to Phytoplankton

Biological Regulatory Networks (BRNs) are interesting for simulation, validation and prediction of biological behaviors. However, from the growing set of available data, it is necessary to build such BRNs beforehand. Given the size of the data, automatic machine learning approaches are handy. Several approaches exist, but there is a special interest in explainable approaches, as they allow to not only make predictions, but also to understand the mutual influences between components.

This presentation will focus on Learning From Interpretation Transition (LFIT), a framework that aims at learning the structure of a BRN by observing its dynamical behavior. LFIT provides explainable predictions in the form of logical rules of the form a_i <- b_j, c_k, d_l… meaning that if b_j, c_k, d_l… are present in the current state, then a_i might be present in the next state. This possibility actually depends on the chosen semantics, which will be explained too.

An ongoing case study will also be presented: the inference of influences between micro-algae species in the field of marine ecology. The variation of population in these species can have an impact on ecological niches, the economy or the climate. The objective of this work is to automate the creation of BRNs of mutually interacting populations using LFIT in order to help put into light biotic interactions, that is, interactions between different species (competition, cooperation…).