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

Ab initio protein structure prediction
Model selection

Ab initio modeling typically generates A large number of structural decoys. A major challenge in this modeling is selecting appropriate models whose structures closely resemble the Native State of the protein, which has given rise to a new research field known as Structure/172.html">Model quality Assessment Programs (MQAP) (Fischer 2006). In general, Selection approaches in modeling can be divided into two main categories: energy-function-based and free-energy-function-based. Energy-based approaches employ various specific potentials, with the state of lowest energy taken as the final structural prediction. In free-energy-based approaches, the Free energy of a given conformation R can be written as

Class="center">

where Z(R) is the restricted partition function, which is proportional to the frequency of occurrence of structures near R during the simulation. It can be estimated using a clustering Procedure with a given RMSD cutoff value (Zhang and Skolnick 2004c).

Among the multitude of free-energy-based Model Selection Methods, this section discusses three energy/scoring Functions: 1) a Rational energy function; 2) an Empirical energy function; and 3) a scoring function that describes the consistency between the target sequence and the model structures. Another popular model quality assessment method relies on consensus Conformations derived from predictions generated by various algorithms (Wallner and Elofsson 2007). This class of methods is also known as meta-servers (Ginalski et al. 2003a; Wu and Zhang 2007). These methods share a conceptual similarity with clustering, as both assume that the most frequently occurring state is the closest to the native one. This approach is primarily used to select models generated by web servers via protein threading (Ginalski et al. 2003; Wallner and Elofsson 2007; Wu and Zhang 2007).