The Structural Role of Feed-Forward Loop Motif in Transcriptional Regulatory Networks

We present multiple approaches to identify the significance of topological metrics that contribute to biological network robustness. We examine and compare the communication efficiency of transcriptional networks extracted from the bacterium Escherichia coli and the baker's yeast Saccharomyces cerevisiae using discrete event simulation based in silico experiments. The packet receipt rate is used as a dynamical metric to understand information flow, while unsupervised machine learning techniques are used to examine underlying relationships inherent to the network topology. To this effect, we defined sixteen features based on structural/topological significance, such as transcriptional motifs, and other traditional metrics, such as network density and average shortest path, among others. Support vector classification is used with these features after parameters were identified using a cross-validation grid-search method. Feature ranking is performed using analysis of variance F-value metric. We found that feed-forward loop (FFL) based features consistently show up as significant in both the bacterial and yeast networks, even at different noise levels. We then use a supervised machine learning technique (random forests) to investigate the structural prominence of the FFL motif in information transmission using subnetworks (larger sample size compared to the unsupervised approach) extracted from Escherichia coli transcriptional regulatory network. Further, we study the role of FFLs in signal transduction within the complete Escherichia coli regulatory network. Although our work reveals a minimal role of FFLs in signal transduction, it highlights the structural role of FFLs in information transmission captured by random forest regression. This work paves the way to design specialized engineered systems, such as wireless sensor networks, that exploit topological properties of natural networks to attain maximum efficiency.