Predictive value, CoMFA models

To test the applicability of this approach, eight compounds were randomly removed. The remaining 52 compounds were used to develop a CoMFA model. The resultant model was then used to predict the solubility of the removed compounds. Results (Table 3.7) show that predicted values agree well with the experimental values. 

The q2 value of a CoMFA model, together with other statistical information from the pis analysis, provides information on the predictive capability of the model. In this study we have generated CoMFA models that describe the pharmacophore either with or without the involvement of hemin, both of which provide good q2 values. Selection of the model that most accurately depicts reality is not trivial since many variables are inherent in the cell-culture bioassay results. However, it may be... 

The predictions from the achiral CoMFA model 199 (2 A/C.3), which includes the racemates, best reflects experimentally derived antimalarial activities. The predicted value for the natural enantiomer of 421 was 110% of artemisinin and 12%... 

Table 20. Prediction Values for 421 and 422 in Different CoMFA Models...

In 1993, Martin and colleagues performed a follow-up study on the same 15 compounds using CoMFA. CoMFA does not produce a simple equation in the manner of Hansch analysis, but CoMFA does calculate activities for the training set compounds. Comparing the predicted and experimental activities allows determination of the model s correlation coefficient. The CoMFA performed by Martin gave an r-value of 0.96. Martin s CoMFA model accounted... 

The root mean square (RMS) of the differences between the experiment and predicted values for the Hansch equation was 0.75, while the CoMFA gave 0.65. A smaller value for the RMS of the differences indicates CoMFA afforded a more accurate model. 

Martin s expanded CoMFA model gave an r-value of 0.92. Six new analogues of 12.18 were prepared, and their activities were predicted with the new model. The RMS of the experimental and calculated activity differences was only 0.36. 

Diaryloxymethano -phenanthrene derivatives 37 CoMFA and CoMSIA CoMFA (q2 = 0.625) and CoMSIA ( 2 = 0.486) models and 7-compound external test set with very good predictive value Shagufta et al. (41)... 

Table 13.5 summarizes the priority setting results for the two data sets using the NCTR Four-Phase system. When only the Phase I and II protocols are used, the system dramatically reduced the number of potential estrogens by some 80 to 85%, demonstrating its effectiveness in eliminating these most unlikely ER binders from further expensive experimentation. The Phase III CoMFA model further reduces the data size by about 5 to 10%. More importantly, the quantitative binding affinity prediction from Phase III provides an important rank-order value for priority setting. 

Several CoMFA and CoMSIA studies were performed, both for optimizing selective androgen receptor modulators and for toxicity predictions. An early AR CoMFA model was reported by Waller et al. [118]. It was based on 28 structurally diverse chemicals from which 21 served as training set and yielded a cross-validated q2 value of 0.792. From the seven test compounds six were predicted within one order of magnitude to the experiment, the remaining deviated by 1.8 log units in K, from the experimental value. 

The last step in a CoMFA study is a partial least squares (PLS) analysis (chapter 5.3) to determine the minimal set of grid points which is necessary to explain the biological activities of the compounds. Most often good to excellent results are obtained. However, the predictive value of the model must be checked by cross-validation if necessary, the model is refined and the analysis is repeated until a model of high predictive ability is obtained. 

QSAR 8 mentioned earlier was derived for the same dataset [75]. Ionization potential (IP), ClogP, molar volume (as MV/lOO) and an indicator variable (1) were used to model the activity. A parabolic model with ClogP unfolded the ideal ClogP value (ClogPo = 5.697) for maximum anti-protease activity. It is of note that the activities of the test set compounds (total eight) were predicted well both by the 2D-QSAR model (r = 0.746) of Wilkerson et al. [75] and the CoMFA model (r = 0.971) of Debnath [99]. 

The number of significant PLS components is established by testing the significance of each additional dimension (PLS component). This is done to avoid overfitted QSARs, which may exhibit lesser, or no, validity. The optimal number of PLS components to be used in conventional analyses is typically chosen from the analysis with the highest cross-validated value, and for component models with identical values, the model having the smallest standard error of prediction, PRESS (see also the following section). Unlike spectroscopic data, where a PLS model typically has more than 10 components, models in 3D-QSAR tend to exhibit less complexity. As a rule of thumb, two to four components should suffice when CoMFA standard fields are used." ... 

Two variable selection methods described in the introduction to PLS, the r2-guided region selection method and GOLPE, have been applied to CoMFA, but only a few direct comparisons are available. GOLPE variable selection - has led to PLS models with higher cross-validated r values but no more accurate predictions. Comparison of models derived from traditional PLS with those using domain-selected or GOLPE-selected variables shows improve-... 

The results of CoMFA and CoMSIA analysis are summarized in Table 7.1. The CoMFA PLS analysis yielded a high cross-validated correlation coeflftcient (f of 0.872 with standard error of prediction SEP of 0.383. The non-cross-validated PLS analysis gave a conventional of 0.974 with SE of 0.172. These values indicated a good statistical correlation and reasonable predictability of the CoMFA model. 

---