Assessing distance measures for protein structure comparison

Zikai Wu,Yong Wang,Enmin Feng,Luonan Chen
2007-01-01
Abstract:Distance measure is a key component for fast protein structure comparison. The identical representation of protein structure combining with different distance measures results in different structure comparison methods. In this paper, we provide and further analyze three structure comparison methods with contact vector representation based on three different distance measures respectively, i.e. Euclidean distance, cross entropy and FDOD function, in an empirical manner. Relying on a public data set, we evaluate the ability for detecting function similarity of the three methods. The comparison results reveal that two information-based measures outperform the Euclidean distance, in particular the method based on FDOD function is the best in the three distance measures. We show that the information-based measure can really identify the subtle difference given abstract representations of protein structures. Moreover, the FDOD for protein comparison is also superior to cross entropy in terms of mathematical consistency.
What problem does this paper attempt to address?