Research Article
Feature ranking in transcriptional networks: Packet receipt as a dynamical metric
@INPROCEEDINGS{10.4108/icst.bict.2014.257930, author={Bhanu Kamapantula and Michael Mayo and Edward Perkins and Ahmed Abdelzaher and Preetam Ghosh}, title={Feature ranking in transcriptional networks: Packet receipt as a dynamical metric}, proceedings={8th International Conference on Bio-inspired Information and Communications Technologies (formerly BIONETICS)}, publisher={ICST}, proceedings_a={BICT}, year={2015}, month={2}, keywords={biological robustness transcriptional networks feature ranking}, doi={10.4108/icst.bict.2014.257930} }
- Bhanu Kamapantula
Michael Mayo
Edward Perkins
Ahmed Abdelzaher
Preetam Ghosh
Year: 2015
Feature ranking in transcriptional networks: Packet receipt as a dynamical metric
BICT
ACM
DOI: 10.4108/icst.bict.2014.257930
Abstract
Machine learning techniques may be useful in determining the features contributing to some biological properties, such as robustness, which is the tendency for biological systems to resist a change of state. In this work, we compare transcriptional subnetworks extracted from the bacterium Escherichia coli and the baker’s yeast Saccharomyces cerevisiae using in silico experiments. We use the packet receipt rate as a metric to quantify biological robustness, which is different from the usual structural metrics since it captures the dynamic behavior of the network. We define seventeen features based on structural significance, such as transcriptional motifs, and conventional metrics, such as average shortest path and network density, among others. Feature ranking is performed, based on a grid-search method to identify Support Vector Machine classifier parameters using cross validation. Our results indicate that feed-forward loop based features are important for bacterial transcriptional networks, whereas network density, degree-centrality based and bifan-based features are found to be significant for yeast-derived transcriptional networks. Interestingly, results suggest that feature significance varies with network size (number of nodes). As a first, this study quantifies the impact of the feed-forward loop and bifan transcriptional motif abundance observed in natural networks.