Two Challenging Difficulties of Protein Complex Prediction

Abstract

A protein complex is a group of proteins which carries out particular functions in the cell. The component proteins of a protein complex are connected via weak physical contacts, called protein–protein interactions (PPIs). Proteome-wide PPIs are determined by high-throughput assays. Thus, it is interesting to computationally predict protein complexes from such PPIs. In this paper, we describe two challenging difficulties of the problem. The first difficulty is that the smallest protein complexes are of size two. It is quite difficult to predict them due to their simple inherent structure. The second difficulty is that some known complexes are overlapped with each other, because it is not trivial how to model such overlaps mathematically. For these issues, we have proposed our own approaches. In both methods, we design a scoring function and algorithms based on Markov chain Monte Carlo to optimize the scoring function. In this article, we briefly show our key regularization terms included in the whole scoring function.