Alignment problem CoMFA

The biggest limitation of the CoMFA method is the alignment step. The algorithm superimposes the portions of the inhibitors that are of similar stmcture, assuming that they bind with similar orientations in the active site of the enzyme, which is not necessarily the case. Also, because of a problem with alignment, a CoMFA may fail when a few molecules are very dissimilar from all others in the series. Like QSAR, CoMFA does not require a stmcture of the relevant biological receptor, but does require knowledge about a series of inhibitory compounds. 

CoMSIA (comparative molecular similarity index analysis) is a recent development from CoMFA and does not suffer from the alignment problem. It has been used to model hERG potassium channel inhibition by drugs [59] and the toxicity of phenylsulfonyl carboxylates [60], organophosphates [61], and polybrominated diphenyl ethers [62], with results comparable to those from CoMFA. 

Figure 17.4 Illustrating the alignment problem in COMFA methods...

With respect to CoMFA, the Compass method effectively reduces the number of descriptors, performing a physico-chemically based variable reduction and overcomes the problem of guessing the best conformation and alignment of the molecules. 

The G-WHIM approach integrates the information contained in —> WHIM descriptors and overcomes any problem due to the alignment of the different molecules and the explosion of variables arising from traditional grid-based QSAR techniques, such as GRID and CoMFA. 

Is 3D-QSAR best left to experts, or can less skilled scientists apply the methods The discussions above may indmidate a nonexpert who had contemplated trying 3D-QSAR methods. In fact, the common approaches such as CoMFA are easily learned and hard to misuse. A little time spent learning how to interpret PLS statistics is all that is needed to supplement the individual s molecular modeling experience. If problems in alignment, choice of conformation, calculation of properties, or suspected nonlinearities arise, an expert collaborator should be sought. In our experience, most data sets fairly easily yield predictive CoMFA models, and those that do not often fail to improve with additional changing of parameters. 

CoMFA is a powerful 3D QSAR method which has already shown its practical value in many cases. The results obtained depend a lot on the care that is taken in the definition of the 3D pharmacophore and in the alignment of the molecules. Soft fields seem to be better than hard fields, especially for the interpretation of the contour maps. Variable selection seems unnecessary if such fields are used. On the other hand, the new GOLPE-guided regional selection looks more promising than other variable selection procedures. A general problem is the... 

---