Quality Assessment of Protein Models

When building a protein model, with or without the aid of experimental information, it is oftennecessary to use an independent measure to evaluate the correctness of the model. This is the role ofModel Quality Assessment programs (MQAPs). Different types of MQAPs have been developed duringthe last decades. The goal of all these methods is to assess the quality of protein models. However, thedefinition of quality differs depending on the problem, thus it is always important to consider thespecific problem to be solved when using an MQAP.Traditionally MQAPs are methods that evaluate the quality of a protein model. Until recently mostwork has been focused on the development of methods aimed at detecting the native structures and toseparate these from incorrect models. However, during recent years other types of MQAPs, includingconsensus based MQAP, has increased in importance and today one of the most important uses ofMQAPs is to select the best out of a set of models built by homology or by other methods. Althoughthese two problems clearly are related, there is no guarantee that a method that works well on the firstproblem works well on the other, in particular when all of the plausible models are of low quality.In this chapter we will first discuss how MQAPs have been used in the past and how they are usedtoday and finally we will present an analysis of how MQAPs performed in CASP7 [1].The first use of MQAPs was to detect erroneous models from X-ray crystallography. X-ray basedmodels might be wrong when the resolution of the diffraction data is low, but other types of errors,including tracing a chain backwards through the electron density also occurs. In this case, the numberof residues in disallowed regions in the Ramachandran plot can often be used as an indicator for thequality of the X-ray model. This so called “stereochemical correctness” could for instance havedetected the wrongly built small subunit of Rubisco by Eisenberg and coworkers [2], which was built inthe reverse direction. In the early 1990’s a number of MQAPs were developed to identify wrongly builtmodels using Ramachandran plots and other measured of “stereochemical correctness” as the mainsource of information. The best known of these methods are PROCHECK [3] and WHATCHECK [4].Today with improved refinement methods the “stereochemical correctness” of X-ray models is almostalways very good. Also, with the introduction of the R-free method [5] where a fraction of the data isonly used for testing the need for MQAPs to validate protein models built from crystallographic2