CoMFA statistical requirements

Not all the above-mentioned studies fulfill the necessary statistical requirements, especially proper PLS component selection by cross-validation. The crucial influence of different CoMFA options on the obtained results has been demonstrated [38,1059]. More objective alignment procedures, additional fields (e.g. a better consideration of polarizability), and more efficient variable selection procedures are needed. In addition, further research is necessary to find out which of the different CoMFA options should be selected to obtain stable and reproducible results. Theoretical and practical aspects, methodology and applications, as well as some limitations of CoMFA and related approaches have recently been reviewed in a book on 3D QSAR methods [38],... 

Comparative molecular field analysis (CoMFA) is a modern, powerful extension of the classical QSAR methods that were developed in the 1960s.14 While Hansch analysis is simple to understand and fairly easy for any medicinal chemist to perform, CoMFA requires specialized software and an understanding of statistics. Since CoMFA is outside the experience of most synthetic chemists, pharmaceutical companies have dedicated computational chemistry groups to handle advanced QSAR tasks. 

A further complication of the interpretation of contour maps results from the requirement of electroneutrality of a molecule. Electrostatic coefficient contours may be found close to ligand atoms that have polar interactions with specific receptor atoms. However, electrostatic contours, of the opposite sign, may also surround ligand atoms not involved in any polar interaction. The charge distribution pattern is felt by CoMFA from both extremes of the molecular dipole. As a result, both elearon-rich and electron-poor regions will be included in the model. Statistically, this is a simple case of correlation between descriptors (collinearity). 

Figure 2 Structural alignment methods with continuous properties. Methods are based on continuous properties using the algorithms implemented in MSA (I), HASL (II), and CoMFA (III). A conformational analysis (optional) can be used to select candidate conformations ( bioactive conformation, or alternatively a set of biologically relevant conformations that are used for the alignment). The structure representation is based on molecular shape (I), a four-dimensional lattice (II). or molecular fields (III). The different algorithms for the alignment require that a reference be cho.sen so that comparisons between each ligand and the reference can be made. A statistical measure of molecular similarity is performed to identify the alignment that maximizes the molecular similarity between the ligands in term.s of three-dimensional overlap (MSA, HASL) or electrostatic potential distributions...

---