Biomimetic Reverse Engineering of
Gene Regulatory Networks

Technologies to assay gene expression levels and protein concentrations are currently advancing at a fast pace, opening the door for reverse engineering of gene networks (see box in the sidebar) using biologically plausible, nonlinear models where traditional methods often fail.

Our approach is based on a genetic algorithm that employs a novel biomimetic representation called Analog Genetic Encoding (AGE), which can be used with nonlinear gene models where analytical approaches or local (gradient-based) optimization methods are not appropriate.

Analog Genetic Encoding allows simultaneous inference of network structure (size, topology) and numerical parameter values. It differs from other state-of-the-art genetic algorithms and global optimization methods in the sense that it mimics the encoding of natural gene regulatory networks. We hypothesize that this is an effective way of incorporating prior biological knowledge in the search.

DREAM2 challenge

DREAM (Dialogue for Reverse Engineering Assessments and Methods) is an initiative to foster the interaction between experiment and theory in cellular network reverse engineering. For the second DREAM conference (DREAM2), a series of reverse engineering challenges have been organized.

We have participated in the DREAM five-gene network challenge. For this challenge, a synthetic-biology gene network was constructed and transfected to an in-vivo model organism. The topology of the network was unknown to the participating teams. The goal of the challenge was to predict the topology of the gene regulatory network from a time-series dataset of q-PCR gene expression measurements.

Our method was acknowledged best performer on this challenge. For details, see the DREAM results page (teams are anonymized, we are team 55).

Relevant publications

Publications are listed on the overview page.

The two companion papers [3,4] describe the best performing method and results of the DREAM2 five-gene network challenge. Ref. [3] introduces the biomimetic reverse engineering approach and describes its application to the challenge. In ref. [4] we show how ensemble methods can be used to analyze a set of inferred networks. We apply these ensemble methods to the results of ref. [3], which allows us to significantly boost the accuracy of predictions in the DREAM challenge.

Ref. [1] reports preliminary results of the biomimetic method on an in silico test case.