Protein Structure and Function: Application of Bioinformatics Methods - John Rigden 2014

ab initio Protein Structure Prediction
Comments and Discussion

Structure/4.html">ab initio Cell/13.html">Protein Structure Prediction, relying solely on an Amino Acid Sequence, is widely considered the “Holy Grail” of protein modeling (Zhang 2008), as a breakthrough in these Methods would mean the ultimate and definitive solution to the problem. Beyond generating spatial structures, ab initio modeling can help us understand the fundamental principles governing natural protein folding. This challenge can be addressed independently of comparative modeling approaches, where a 3D structure is built by copying the backbone framework of previously solved structures.

An ideal ab initio modeling scheme would be an experiment treating protein atoms as interacting particles, with interactions governed by an exact physical potential, and the folding process simulated by solving Newton's equations of motion at every step. Several such molecular dynamics (MD) simulations have been performed using the classical CHARMM and AMBER force fields. Although MD simulation is an indispensable tool for studying protein compaction, its success in structure prediction remains quite limited. One reason is the immense computational power required to simulate even medium-sized Proteins. On the other hand, empirical (as well as hybrid, knowledge-based, and empirical) methods are advancing rapidly; numerous Examples now demonstrate successful low- to medium-resolution models of proteins up to 100 residues, often featuring the correct topology. Furthermore, albeit very rarely, high-resolution models (under 2 Å for C_a atoms) have been successfully reported (Bradley et al. 2005).

Modern ab initio methods for protein structure prediction typically incorporate the maximum Amount of Information from known structures. There are several reasons for this. First, utilizing local structural fragments extracted directly from the PDB database helps reduce the degrees of freedom and conformational search Entropy, while preserving the accuracy of native protein structures. Second, employing an empirical potential derived from extensive statistical data on known structures helps maintain the delicate balance of complex interplay among various energy term contributions (Summa and Levitt 2007). These empirical potential terms are rigorously parameterized. Thanks to recent advances in conformational search algorithms, many computational and procedural steps have become semi-automated. For these reasons, the accuracy of ab initio methods applied to proteins of 100–120 residues has improved significantly over the past decade.

Further progress requires concurrent advancements in both accurate potential energy Functions and efficient optimization techniques. This means that independent research and development of potential energy functions remain crucial; simultaneously, regular benchmarking of various conformational search methods is essential to independently evaluate both the advantages and limitations of currently available search strategies.

It is important to note that ab initio Prediction Methods Based strictly on physicochemical interaction principles currently lag significantly behind bioinformatics and template-based modeling methods in terms of both accuracy and speed. Nevertheless, physical atomic potentials have proven valuable for refining the packing details of side-chain and backbone atoms. Consequently, developing sophisticated methods that combine both empirical and physical energy potentials likely represents a promising approach to advancing ab initio modeling.

Acknowledgements. This work was supported in part by the KU Start-up Fund 06194, the Alfred P. Sloan Foundation, and NIH National Institute of General Medical Sciences Grant R01GM083107.