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

Ab initio protein structure prediction
Model selection
Rational energy function

To develop all-atom Structure/4.html">Rational Energy Functions, Lazaridis and Karplus (1999a) utilized the solvation potentials of the CHARMM19 (Neria et al. 1996) and EEF1 (Lazaridis and Karplus 1999b) force fields, using these values to discriminate between native protein structures and decoys generated by threading sequences through alternative protein backbones. It was found that the energy of the Native State was, in most cases, lower than that of the decoys. Later, The Development of rational energy functions incorporated the CHARMM force field alongside the continuum solvent model (Petrey and Honig 2000), the CHARMM force field with the GB method (Dominy and Brooks 2002; Feig and Brooks 2002), OPLS with GB (Felts et al. 2002), the AMBER force field with the GB method (Lee and Duan 2004), and the AMBER force field combined with the Poisson-Boltzmann solvation potential applied to several sets of structural decoys—including those from Park and Levitt (Park and Levitt 1996), Baker (Tsai et al. 2003), Skolnick (Kihara et al. 2001; Skolnick et al. 2003), and CASP (Moult et al. 2001)—as reported by Hsieh and Luo (2004). All these authors obtained consistent results: the energy values calculated using these potentials were lower for native protein structures than for decoys. It might seem that the successful application of physical potential-based discrimination models contradicts findings from other, less successful physical Methods. However, it was recently demonstrated that combining AMBER force field potentials with the GB method can only distinguish native structures from coarsely minimized decoys generated by TASSER (Zhang and Skolnick 2004a; Wroblewska and Skolnick 2007). Following a 2 ns Molecular Dynamics (MD) simulation of the decoys, none of the native structures exhibited an energy lower than the minimum energy of the decoys, and the correlation between energy and RMSD was close to zero. These findings partially resolve the discrepancy between the widely reported successes of decoy discrimination using physical potentials, on the one hand, and the less successful outcomes of protein backbone packing and refinement methods, on the other.