Multi-scale Modelling of Biomolecules

Abstract

Descriptions of large biomolecules, especially RNA and proteins, typically either focus on very localized descriptors like dihedral angles, which fail to adequately describe large scale structure, or employ a landmark shape treatment, which favors large scale properties and misses local features like fairly rigid atomic bond lengths and angles. To facilitate clustering on the microscale we have developed a torus PCA method applicable to dihedral angles. On a somewhat larger scale, we model the molecule backbone shape by landmark size-and-shapes. Using this description on multiple scales, we propose an approach to correcting possibly poor fits in RNA structures.