Algorithm Engineering for Color-Coding to Facilitate Signaling Pathway Detection

Abstract

To identify linear signaling pathways, Scott et al. [RECOMB, 2005] recently proposed to extract paths with high interaction probabilities from protein interaction networks. They used an algorithmic technique known as color-coding to solve this NP-hard problem; their implementation is capable of finding biologically meaningful pathways of length up to 10 proteins within hours. In this work, we give various novel algorithmic improvements for color-coding, both from a worst-case perspective as well as under practical considerations. Experiments on the interaction networks of yeast and fruit fly as well as a testbed of structurally comparable random networks demonstrate a speedup of the algorithm by orders of magnitude. This allows more complex and larger structures to be identified in reasonable time; finding paths of length up to 13 proteins can even be done in seconds and thus allows for an interactive exploration and evaluation of pathway candidates.