Carnegie Mellon Research Ties Biological and Computing Networks

• By Dian Schaffhauser
• 05/01/14

A team of Carnegie Mellon University (CMU) researchers is using a biological model to understand how computer networks might better respond to security assaults. The scientists are taking a clue from the humble yeast cell as it protects itself against environmental threats.

According to the scientists, a yeast cell has about 6,000 genes, 20 percent of which are considered "essential." That means that if one or more of those particular genes are removed (as in DNA experimentation), the cell will die. However, the removal of those particular genes doesn't often happen in nature. They're protected by other genes lying nearer the cell surface that have evolved to withstand varying levels of stress, otherwise known as environmental "noise."

The research suggests that the topology of molecular subnetworks or modules is "tightly linked" to the level of noise the module expects to encounter. Those less exposed are more vulnerable than those more exposed.

The CMU team is looking into how that evolutionary approach of cells could be applied to computational networks that need to be designed to withstand stresses of their own — such as cascading failures from an overloaded power grid or some kind of cyber-attack.

According to Saket Navlakha, a post-doctoral researcher in CMU's Machine Learning Department, frequently, network security is evaluated by removing a server to see how the network responds. Maybe this isn't always the right approach. The yeast cell connection suggests that many network failures "involve the loss of multiple, neighboring nodes."

Communication networks, the team suggested, could be "tailored" for the environment in which they're operating. For example, highly connected networks running the Internet or its large service providers may be efficient and fast, but they're also open to a greater number of infections and cascading failures; peer to peer networks are connected more "sparsely" and may be less efficient, but they also make up networks that can better tolerate attacks.

The researchers have created an algorithm that can evaluate the topology of a network to determine its suitability for different environments. A tightly restricted military network, for example, doesn't have to worry too much about malevolent viruses or noise since access to it is so closely monitored. However, a wireless sensor network deployed over a wide area might need to be designed to sustain continual loss of random nodes.

Likewise, noted team member Ziv Bar-Joseph, an associate professor in CMU's Machine Learning Department and the Ray and Stephanie Lane Center for Computational Biology, the role of "external noise" in biology might turn out to be an insight of use in understanding how biological networks function.

"Over the last few years we have [begun] to witness a change in how biologically inspired computational methods are derived and studied," he wrote in a grant application. "A number of recent bi-directional studies, by us and others, have demonstrated that thinking computationally about the settings, requirements and goals of information processing in biological networks can both improve our understanding of the underlying biology and lead to the development of novel computational methods providing solutions to decades-old problems.

Future research will look at biological processes of flies and E. coli as well as computation processes such as distributed computing and machine learning.

The findings of the team have been published in the latest issue of the Journal of the Royal Society Interface. It is funded by grants from the National Institutes of Health and the James S. McDonnell Foundation.