Predicting Protein-Protein Interactions

Source: Najafabadi and Salavati (2008). Sequence-based prediction of protein-protein interactions by means of codon usage. Genome Biology 9:R87

My dear friend Hamed, now working in the Parasitology department of McGill University, has published a recent paper in Genome Biology which (believe it or not ;)) has made the highly accessed set in this rather prestigious journal. I asked him to write a summary of his paper and despite his tight schedule (believe me, it's TIGHT), he sent me the following. ENJOY.

The mainstay research in our lab is focused on trypanosomatids, human parasites with almost-newly sequenced genomes. Their distant homology with well-studied organisms such as yeast, bacteria and animals makes it extremely difficult to draw any conclusions regarding the functions of the majority of their genes using homology-based methods. Although this paper does not even mention the trypanosmatids, it describes a method that we basically developed for prediction of functional linkages between proteins in these parasites. We surprisingly found that codon usages of the proteins that either physically or functionally interact with each other share more similarities than the codon usages of unrelated proteins. This may be due to factors such as the requirement for similar expression levels of interacting protiens, physical proximity of the genes of the interacting proteins on genome, or unknown evolutionary forces yet to be determined. We developed a classifier that uses codon usage for prediction of physical/functional interactions of proteins. Most notably, we found that combining this classifier with previously published classifiers almost doubles their sensitivity. We also showed that, no matter which organism, once a suitable training set is available this method is the first choice for homology-independent prediction of functional and physical interactions.