Internal predictivity

On the basis of their evaluation and our internal predictive VolSurf model [160] for this series (r 0.81, q 0.60, 4 PLS components), it can be concluded that factors like size and shape, which had previously been reported to affect paracellular permeability, are indeed important in the VolSurf PLS model to explain the local structure-permeability relationship of one particular scaffold. Hence, local statistical models provide a qualitative ranking of candidates, and thus are valuable for optimization of pharmaceutically relevant compounds, especially if combined with additional models to understand affinity, selectivity or any particular pharmacokinetic behavior. 

The published QSAR [59-61] and 3D-QSAR [62-65] models for HDAC inhibitors were used to explain the differences in activity of hydroxamate-based compounds. All the reported models, which showed moderate to good internal predictivity, were mainly used in a retrospectively way to explain the biological activities of H DAC inhibitors. Generally, the 3D-QSAR models were compared with ligand docking results to get insight into the structural requirements for anti-HDAC activity. 

If formulations with three or more release rates are used to develop the IVrVC model, no further evaluation beyond this initial estimation of prediction error may be necessary for non-narrow therapeutic index drugs (Category 2 a and b apphcations, see page 12). However, depending on the results of this internal prediction error calculation, determination of prediction error externally may be appropriate. 

If these criteria are not met, that is, if the internal predictability of the IVIVC is inconclusive, evaluation of external predictability of the IVIVC should be performed as a final determination of the ability of the IVIVC to be used as a surrogate for bioequivalence. 

With the exception of narrow therapeutic index drugs, the external predictability step in the IVIVC evaluation process may be omitted if the evaluation of internal predictability indicates acceptable % PE. However, when the evaluation of internal predictability is inconclusive, evaluation of external predictability is recommended. 

Predictability Verification of the model s ability to describe in vivo bioavailability results from a test set of in vitro data (external predictability) as well as from the data that was used to develop the correlation (internal predictability). Percent Prediction Error % PE = [(Observed value - Predicted value) Observed value] X 100... 

To circumvent this issue, cross-validation methods have been proposed to evaluate the internal predictivity of the model by discarding one or several com-... 

WRF/Chem Sulfate, nitrate, ammonium, BC, OC, water in all 3 aerosol modules, sea-salt, and carbonate in MOSAIC/ MADRID Modal (3) variable (MADE/ SORGAM) Sectional (8) variable (MOSAIC/ MADRID) single size distribution Internal Predicted/Predicted similar to MIRAGE2 similar to MIRAGE2... 

Internal predictability is established for each formulation used to develop the IVIVC model. Based on in vitro data, the IVIVC relationship is used to recalculate the initial in vivo curves, and then the predicted bioavailability is compared to the observed bioavailability for each formulation and a determination of the error prediction is made. Internal predictability is acceptable when the average percent prediction error is below 10% for Cmax and AUC, and none of the formulations have a prediction error greater than 15%. If the results are not acceptable, then external formulation with new test data are needed corresponding to an external predictability process. 

Since the IVIVC model is going to be used to predict the plasma concentrationtime profile, it is imperative to assess the predictive performance of the model via the assessement of the prediction error of the model. Depending on the intended application of the IVIVC and the therapeutic index of the drug, evaluation of the internal or external predictability may be warranted. Evaluation of internal predictability is based on the data that was used to develop the IVIVC. Evaluation of... 

If the IVIVC for a non-narrow therapeutic index drug was developed with formulations with three or more release rates, the evaluation of the internal predictability would be sufficient to determine its appropriateness. 

For internal predictability, an average absolute prediction error of less than 10% for both AUC and Cn,ax establishes the predictive abiUty of the IVIVC. In addition, the percent error for each formulation should not exceed 15%. If the above criteria are not met, the IVIVC is declared inconclusive and in this case the evaluation of the external predictability of the IVIVC is required. 

This evaluation may be performed either by use of the same data, as included in the establishment of an IVIVC (internal predictability), or by use of other data sets (external predictability). The criteria for concluding a level A IVIVC in a regulatory context requires, in the case of internal predictability, an average percentage PE of < 10 percent for Cmax and AUC, respectively, and the percentage PE for each formulation regarding these two bioavailability variables should not exceed 15 percent (FDA 1997a). 

Statistical parameters of CMF, CoMFA and CoMSIA models are shown in Table 13.3. They include the values of 4 statistical parameters and RMSE characterizing internal predictive performance, and RMSE —external predictive... 

External Prediction Parameters Internal Prediction Parameters ... 

The problems of this data set are easily understood if a Free-Wilson analysis is applied. " The training set compounds ( 1-21) can be described by a simple one-parameter regression equation (equation 1 the term 4,5-C=C- indicates the presence or absence of a cycloaliphatic 4,5-double bond in ring A of the steroids). The internal predictivity of this model (Q = 0.726 spress = 0.630) and the test set predictivity (n = 10 = 0.477 spress = 0.733) are even slightly... 

Despite the worse fit and internal predictivity, as compared with equation (1), the validity of this model is proven by its excellent test set (compounds 13-22) predictivity pnsd = 0.909 spRESs = 0.406). The differences between both models, especially in their test set predictivity, provide striking evidence for the influence of the training and test set selections on the obtained results. Thus, a careful selection of the training set molecules is of utmost importance. A broad variety of structural features should be included in these molecules, in order to allow reliable predictions for the test set compounds. 

Statististical data for fit and internal predictivity should be given. 

Despite the worse fit and internal predictivity, than equation (28), this model is especially suited for test set prediction (compounds 13-22 n 10 = 0.909, spress =0.406)... 

---