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

Ab initio protein structure prediction
Model selection
Empirical energy function

In 1990, Sippl, utilizing statistical data on Proteins of known Structure from the PDB database, developed a residue-residue pairwise interaction potential (Sippl 1990); the latest version of the method is PROSA II (Sippl 1993; Wiederstein and Sippl 2007). Since then, numerous Empirical potentials have emerged, including atomic interaction potentials, solvation potentials, Hydrogen bond potentials, torsion angle potentials, and others. In coarse-grained potentials, each residue is represented either by a single atom or by multiple atoms. For instance, there are potentials based on C_a atoms (Melo et al. 2002), C_ß atoms (Hendlich et al. 1990), side-chain centers of mass (Bryant and Lawrence 1993; Kocher et al. 1994; Thomas and Dill 1996; Skolnick et al. 1997; Zhang and Kim 2000; Zhang et al. 2004), and side-chain centers of mass combined with C_a atoms (Berrera et al. 2003). One of the most widely used potentials, RAPDF, is all-atom, accounts for residue-specific features, and is distance-dependent (Samudrala and Moult 1998). The potential calculation takes into account the distance between 167 specific amino acid pseudoatoms. Later, other atomic potentials with various reference states were introduced, including KBP (Lu and Skolnick 2001), DFIRE (Zhou and Zhou 2002), self-RAPDF (Wang et al. 2004), VICTOR/FRST (Tosatto 2005), and DOPE (Shen and Sali 2006). According to their developers, utilizing these potentials allows for the identification of the native Cell/13.html">Protein Structure among decoy sets. However, The problem of selecting models close to the native structure from a pool of decoys in these Methods remains unsolved (Skolnick 2006). This challenge is arguably more critical than recognizing the native structure, given that, to date, there are virtually no protein molecules whose structures have been successfully determined solely by computational modeling methods. According to the CAFASP4-MQAP 2004 experiment results (Fischer 2006), the Victor/FRST (Tosatto 2005) and MODCHECK (Pettitt et al. 2005) Energy Functions demonstrated the highest performance. The Victor/FRST function incorporates an all-atom pairwise interaction potential, a solvation potential, and a hydrogen bond potential. MODCHECK includes a C_ß atom interaction potential. In the CASP7-MQAP competition in 2006, the Peons method, developed by Elofsson's group (Wallner and Elofsson 2007), showed the best performance.