Carnegie Mellon scientists unveil new tool to understand evolution of multi-domain genes

Neighborhood Correlation exploits the structure of a statistically weighted sequence similarity network to differentiate multi-domain genes with shared ancestries from multi-domain genes that result from domain shuffling. Gene duplication creates a specific signature in the network, while domain insertion creates a different characteristic signature. Neighborhood Correlation captures these signatures, giving pairs that arose through duplication, and hence share common ancestry, a higher score than genes that share an inserted domain, but not a common ancestor.

The Carnegie Mellon scientists tested Neighborhood Correlation against 20 protein families — including Kinases, the largest multi-domain family found in humans — whose ancestral relationships are well established through lab-based research. The tool worked remarkably well in verifying the ancestral patterns of multi-domain gene evolution for these families, much better than the tools we use today, Durand said.

Today’s computational tools use sequence similarity, assuming that genes with similar sequences indicate common ancestry. Those methods also use the length of the similar region to rule out similarity that arose due to inserted domains. They reason that the longer the sequence shared by two multi-domain genes, the more likely that those two genes share a common ancestor.

But Durand’s tests showed that this assumption often does not hold. Her team found disturbing results when they compared sequence similarity to their Neighborhood Correlation method in evaluating the 20 gene families with established histories. The sequence similarity method actually yielded false ancestral associations and missed true ancestral relationships.

Neighborhood Correlation is successful because it takes both gene duplication and domain insertion into account.

“Not only do we show that Neighborhood Correlation works empirically, we also provide a sound evolutionary argument as to why it should work,” Durand observed. “Our results show that the organization of sequence similarity network contains evidence of ancient evolutionary processes. This has exciting implications for future studies. We hope that comparing the sequence similarity networks of different species will reveal how evolutionary processes differ in plants, animals and fungi,” Durand said. “Multicellularity evolved independently in each of those groups. To go from a single cell to many cells acting together, each time nature had to solve the same problems of cellular communication and control. But are the solutions the same in each lineage" How those problems were solved is a fascinating question.”