An Artificial Neural Network Approach to Improving the Correlation Between Protein Energetics and the Backbone Structure
Computational approaches to modeling protein structures have made significant advances over the past decade. However, the current limitation in modeling protein structures is to produce protein structures consistently below the limit of 6 Å compared to their native structure. Therefore, improvement of protein structures consistently below the 6 Å limit using simulation of biophysical forces is of significant interest. Current protein force fields such as those implemented in CHARMM, AMBER, and NAMD have been deemed complete, yet their use in ab initio approaches to protein structure determination has been unsuccessful. Here, we introduce a new approach in evaluation of protein structures based on analysis of energy profiles produced by the SCOPE software package. The latest version of SCOPE produces a hydrogen bond profile that is substantially more informative than a single hydrogen bond energy value. We demonstrate how analysis of SCOPE's energy profile by an artificial neural network shows a significant improvement compared to the traditional force‐based approaches to evaluation of structures. The artificial neural network based analysis of SCOPE's energy profile showed identification of structures to within the range of 1.5–3.0 Å of the native structure. These results have been obtained by testing structures in the same Homology, Topology, Architecture, or Class of the CATH family.
Published in Proteomics, Volume 13, Issue 2, Winter 2013, pages 230-238.
© Proteomics 2013, Wiley
Fawcett, T., Irausquin, S., Simin, M., & Valafar, H. (2012). An artificial neural network approach to improving the correlation between protein energetics and the backbone structure. PROTEOMICS, 13(2), 230-238. doi: 10.1002/pmic.201200330