Seminar in applied mathematics and statistics

SPEAKER: Michael Stumpf

TITLE: Approximate Bayesian Computation and Network Evolution Models

ABSTRACT:
Approximate Bayesian computation (ABC) allows us to tackle inference problems, where the likelihood is intractable. Models of network growth and evolution that can capture the complexity of real biological networks certainly fall into this category. As ABC relies on comparisons of the summary statistics of real and simulated data, we need to define suitable network statistics. I will discuss some popular choices before introducing concepts from spectral graph theory and how these can be applied to real biological networks in an ABC framework. Simplicial complexes, a fairly recent development, provide an opportunity to describe networks in very fine detail and naturally capture the scale-richness of real network, that are absent from conventional network models. I will conclude with a comparison of the evolutionary characteristics of protein-protein interaction networks, and an outlook of the possibilities of using ABC for model averaging for complex datasets.

Tea and chocolate will be served in room 04.3.15 after the seminar.