Datamining Protein Structure Databanks for Crystallization Patterns of Proteins
A study of 345 protein structures selected among 1,500 structures determined by nuclear magnetic resonance (NMR) methods, revealed useful correlations between crystallization properties and several parameters for the studied proteins. NMR methods of structure determination do not require the growth of protein crystals, and hence allow comparison of properties of proteins that have or have not been the subject of crystallographic approaches. One‐ and two‐dimensional statistical analyses of the data confirmed a hypothesized relation between the size of the molecule and its crystallization potential. Furthermore, two‐dimensional Bayesian analysis revealed a significant relationship between relative ratio of different secondary structures and the likelihood of success for crystallization trials. The most immediate result is an apparent correlation of crystallization potential with protein size. Further analysis of the data revealed a relationship between the unstructured fraction of proteins and the success of its crystallization. Utilization of Bayesian analysis on the latter correlation resulted in a prediction performance of about 64%, whereas a two‐dimensional Bayesian analysis succeeded with a performance of about 75%.
Published in Annals of the New York Academy of Sciences, Volume 980, Issue 1, Winter 2006, pages 13-22.
Annals of the New York Academy of Sciences 2002, Wiley
Valafar, H., Prestegard, J., & Valafar, F. (2002). Datamining Protein Structure Databanks for Crystallization Patterns of Proteins. Annals Of The New York Academy Of Sciences, 980(1), 13-22. doi: 10.1111/j.1749-6632.2002.tb04885.x