Protein structures detecting errors

Although comparative modeling is the most accurate modeling approach, it is limited by its absolute need for a related template structure. For more than half of the proteins and two-thirds of domains, a suitable template structure cannot be detected or is not yet known [9,11]. In those cases where no useful template is available, the ab initio methods are the only alternative. These methods are currently limited to small proteins and at best result only in coarse models with an RMSD error for the atoms that is greater than 4 A. However, one of the most impressive recent improvements in the field of protein structure modeling has occurred in ab initio prediction [155-157]. 

During the course of the structure validation project, we have discovered about one class of errors every 2 weeks. Most of these errors are fully unimportant to the average PDB-file user, but they are errors nonetheless. Consequently, it is impossible even to list all types of errors here. In this chapter, we have to limit ourselves to a discussion of a few classes of errors that, if not detected, could severely hamper the scientist who bases an experiment on a three-dimensional structure. We will discuss flipping of asparagine, histidine, and glutamine side-chains, administration of alternate atoms and residues, and the role of water molecules in protein structures. A large class of error types is formed by nomenclature errors. Table 3 lists some nomenclature errors we found in the PDB. 

Water molecules are an essential aspect of protein structures. Without knowledge of the location of all tightly bound waters, many aspects of the structure, function, and stability of proteins cannot be properly studied. Unfortunately, waters are often abused in crystal structure determination. From the kinds of errors we detect, we assume that some crystallographers, shortly before they manually place waters in the density map, use software that places a water molecule close to each alpha carbon (in a well-determined structure there is about 1 bound water molecule visible per residue). Unfortunately, it regularly occurs that a few waters that are not moved to another position are forgotten, and remain part of the structure. 

The difference Fourier technique is also very useful in refinement of protein structures. An analysis of the errors and their treatment is discussed by Henderson and Moffat, who find that a difference Fourier map is able to detect much smaller features of electron density than those revealed by a normal Fourier map with the same phases. 

There is continuous progress in molecular dynamics and energy minimization. The main topics are force field improvement [26,53-56], incorporation of effective solvation term [57], b Initio modeling of small protein [49—5155.59], the incorporation of real data for x-ray refinement [60] and NMR structure determination [61], However, much work still has to be done to improve the force fields used for ensr"1 calculations before sn,r but trivial errors be detected... 

Flack, H. D. Avoidance, detection and correction of systematic errors in intensity data. In Crystallographic Computing 3 Data Collection, Structure Determination, Proteins and Databases. (Eds., Sheldrick, G. M., Kruger, C., and Goddard, R.) pp. 3-17. Clarendon Press Oxford (1985). 

Disordering of the crystal structure is not an error on the part of the crystallographer unless it goes undetected. That said, a disorder presents its own set of practical problems for setup in a virtual screen. Disorder reflects a situation where a part of the underlying molecule can be positioned in alternative locations. If detected, the modeler is effectively presented with two or more different models for the protein in question. The modeler then has to decide whether to dock into one or the other, or dock into both. Ensemble and flexible side chain docking methods are well suited to this problem, as they allow the user to model multiple conformations in a single docking run. 

---