Page 112 out of 225 pages in this book.
Tripos Bookshelf  >  QSAR  >  QSAR Theory  >  CoMFA Theory

4.3.3 CoMFA PLS Runs

In general the procedure for CoMFA is exactly the same as any other SYBYL/QSAR analysis, as described in Multivariate Analyses. Here are discussed some issues resulting from the large number of explanatory variables hidden within a CoMFA analysis.

• Scaling
• Validating the CoMFA Procedure
• Validating Your Results
• Faster Processing
• What if q² is not Good Enough?

4.3.3.1 Scaling

Columns other than CoMFA may be included, such as logP or molar refractivity, which complement CoMFA descriptors by correlating with biotransport phenomena rather than specific receptor interactions. To avoid complications arising from the variance contributed by the numerous CoMFA variables, and to counteract a tendency for the steric CoMFA field to swamp the electrostatic CoMFA field, a default column scaling of CoMFA Standard is used whenever a CoMFA column is included in PLS (Scaling option menu set to CoMFA Standard in the Partial Least Squares Analysis Dialog or via the command TAILOR SET PLS SCALING_METHOD COMFA_STD). CoMFA Standard scaling gives each individual CoMFA field and each other column the same potential influence on the resulting QSAR. (Technically, CoMFA Standard scaling is a variant of Autoscale in which CoMFA variables are affected by the overall field mean and standard deviation, rather than by the individual column mean and standard deviation.)

CoMFA Standard scaling often gives results very different from (and probably better than) those obtained using uniform (None) weighting. In particular, electrostatics may be much more important. You can also supply your own relative weights for different kinds of explanatory variables before beginning the PLS run (Scaling option menu set to Specify in the Partial Least Squares Analysis Dialog or via the command TAILOR SET PLS SCALING_METHOD USER_SPECIFIED); you are shown the CoMFA Standard weights and given an opportunity to change these weights. (A higher value means greater influence.)

CoMFA Standard scaling and Autoscale set weights for all grid points within a field to be the same or allow them to vary independently, respectively. MSS: QSAR >>> PLS Region Focusing... (QSAR COMFA REGION FOCUS) in Advanced CoMFA represents a useful compromise between these extremes. Region focusing uses some parameter drawn from a CoMFA PLS model -- which draws on information from all grid points -- to assign a weight to each grid point. Any grid point-specific PLS parameter may be used, but the most robust are Discriminant Power and Modeling Power. Discriminant Power is that fraction of the variance in a model's components derived from each grid point, whereas Modeling Power is the fraction of the variance at each grid point subsumed by a model.

More information on region focusing.