Prediction quality

The improvement of the quality of a design is seen as the primary need of industry, but to facilitate this we need appropriate methods for predicting quality and evaluating the long-term quality of an engineer s design (Morup, 1993 Russell and Taylor, 1995 Shah, 1998 Taguchi et al., 1989). However, there is relatively little work... 

In many practical problems, interactions between the variables appear so that the absolute global optimum can be found heavily. As an example, wavelength selection in NIR determination of blood glucose (see Sect. 6.2.6) is considered. The aim of the selection is to find such combinations of wavelengths with which calibration models are obtained their prediction quality is as near at the global optimum as possible (Danzer et al. [2001], p 174). The number of combinations C for the selection of k wavelengths from n channels of the spectrometer is given by... 

Many advances have been made in computational ADME modeling. For many ADME properties, models now exist which provide reasonably good predictive quality and can be deployed to aid medicinal chemists in drug discovery projects. 

The predictive quality of the models is judged according to the cross-validated R2, known as q2, obtained using the leave-one-out (LOO) approach, which is calculated as follows ... 

In Ma in PCR. m the PRESS values for all 1 to ne eigenvectors used in U, S and V to compute the predicted qualities qs,cross are stored in a vector PRESS that is displayed in Figure 5-65. The figure does not show a clear minimum. In Figure 5-66 we show the results of the cross-validation for ne=12 this number has already been used for the calibration in Figure 5-60. 

Figure 5-66 shows the relationship between the true and cross-validation predicted qualities. Note the small but significant drop in the correlation compared to pure calibration, as shown in Figure 5-60. Calibration invariably produces a better correlation than prediction.

During cross validation, it is most convenient to compute this prognostic vector as a function of the number of factors. This results in a collection of prognostic vectors, conveniently stored in a matrix V prog. After determination of the optimal number of factors, the appropriate column can be selected as prognostic vector. The predicted quality qs for an unknown sample with the spectrum ys is then calculated as ... 

In view of the similarity of the PRESS results for PCR and PLS, it is not surprising that the predicted qualities are very similar for the two methods if the optimal numbers of factors is used. Figure 5-68 summarises the comparison. 

Outliers or inhomogeneous data can affect traditional regression methods, hereby leading to models with poor prediction quality. Robust methods, like robust regression (Section 4.4) or robust PLS (Section 4.7.7), internally downweight outliers but give full weight to objects that support the (linear) model. Note that to all methods discussed in this chapter robust versions have been proposed in the literature. 

Reconstituted HCE (order no. RHC/S/5) is available from Skinethic, Nice, France. It consists of transformed human corneal epithelial cells immortalized by Roger Beuerman from the Louisiana State University-Eye Center, New Orleans, USA. This reconstituted human corneal epithelium forms a multilayered cell culture model that does not exhibit tight junctions and is therefore unsuitable for in vitro drug transport studies. Main focus of the model is eye irritation and toxicity testing and it is commonly used in this area with good prediction qualities. Major advantage of this model is the high reproducibility and uniform appearance. The commercial availability provides a ready-to-use model that is easy to handle. 

The first QSPR models for skin tried to establish linear relationships between the descriptors and the permeability coefficient. In many cases validation of these models using, for example, external data sets was not performed. Authors of more recent models took advantage of the progress in statistical methods and used nonlinear relationships between descriptors and predicted permeability and often tried to assess their predictive quality using some validation method. 

The model indicated that hydrophilicity negatively impacts the BBB permeation. Further, the most important descriptors were found to be HBAC and MVOL. The third term could be replaced with HBDN with only a minor deterioration in the predictive quality of the model (Eq. 38) ... 

Additional examination of the model s fit is performed through the comparison of the experimental and predicted bioactivities and is needed to statistically ensure that the models are sound. The methods of chi (%) and root-mean squared error (RMSE) are performed to determine if the model possesses the predictive quality reflected in the R2. The use of RMSE shows the error between the mean of the experimental values and predicted activities. The chi value exhibits the difference between the experimental and predicted bioactivities ... 

Table 2. PLS model prediction quality as a function of reaction temperature...

Quantitative models for predicting quality can be classified into two categories (1) fundamental process models, which are based on physical and chemical events that occur in the autoclave, and (2) regression-type models, which are based on a statistical fit of the observed product quality to the input raw material properties and the process conditions used. 

The quality of design (product and its manufacturing process)—the ability to reliably predict quality and performance, process monitoring and controls, process capability and appropriate risk-mitigation strategies—provides an opportunity to achieve real time quality assurance (the ultimate level of efficiency). This also provides an excellent opportunity to develop efficient and effective quality assurance systems as an alternative to market or public standards (18). 

Altogether, the data reported in this section indicate a very good predictive quality of the model simulations this implies in the first place that the SCR kinetics estimated over powdered catalyst were successfully validated at this bigger scale. However, the excellent agreement between monolith data and model predictions based on intrinsic kinetics also confirms the accurate model description of physical phenomena, specifically external and intraporous mass transfer, which were not significant in the microreactor runs over the powdered catalyst, but played an important role in the monolith runs, as pointed out by the direct comparison in Fig. 44. 

As shown, e.g., in Fig. 50, a good correlation between simulation and experiment was apparent in all cases, which confirms the good prediction quality of the SCR converter model also in the presence of N02. It should be noted that the NOx conversion curves of these experiments (see e.g. Fig. 50) reveal a typical behavior. After the start of the urea dosage a steep increase in the NOx conversion rate appears, which is mainly determined by the fast SCR reaction converting all of the available N02. In the following time period, the increase of the NOx conversion is much slower, with the limiting factor being the buildup of stored NH3. 

An alternative method of handling crude oil is the energy refinery in which crude is split into a number of fractions which can be treated by proven processes to yield two products, SNG and low sulfur fuel oil One such scheme is shown m Fig, 3. The advantages of using the CRG process as the final stage m the production of SNG are high efficiency and low capital cost, the predictable quality of the product gas, and the absence of by-products,... 

Finally, the impact of possible calibration instability on the prediction quality must be understood. In this case, spectrograph and laser wavelength calibrations were very stable and easy to update, but of minor concern because of the broad bands being used. However, since the intensity calibration function is non-linear, any changes in it could unequally affect bands used in the calibration ratio and introduce error in the prediction. Newer equipment offers easy intensity calibration routines but this can be difficult to use automatically with immersion probes since it requires that they be removed from the process. 

In Ref. [52] it was demonstrated that experimentally derived structural information such as the existence of S-S bonds, protein side-chain ligands to iron-sulfur cages, cross-links between side chains, and conserved hydrophobic and catalytic residues, can be used by GAs to improve the quality of protein structure prediction. The improvement was significant, usually nudging the prediction closer to the target by more than 2 A. However, even with this improvement, the overall prediction quality was still insufficient, usually off by more than 5 or 6 A from the target structure. This was probably due to the small number and the diverse nature of the experimental constraints. 

The predictive quality and robustness of the model were examined using 16 molecules of the test set defined previously. Figure 9.5 plots the predicted HERG activities of the test set molecules versus experimental measured values, showing good agreement between the two. 

Dependent process variables are those that result from implementing the various independent parameters under direct control during processing. As independent parameters are altered there will be some effect on the process conditions as reflected by the dependent variables. Proper control of independent variables will result in achieving the acceptable dependent variables, assuring a high level of confidence that the finished product exhibits the expected attributes. Adequately defined and executed process parameters would be expected to yield a finished product of predictable quality, purity, efficacy, and stability. 

Results of the Discriminant Analysis. For each of the two set-ups, the following parameters were used peak maximum, onset temperature, heat necessary to melt the unstable, P and p crystals (absolute and relative values). This leads to eight parameters per storage time, which, multiplied by four storage times (0, 1, 4, and 24 h after production), leads to 32 parameters per individual. However, some parameters failed the tolerance test. This means that for that parameter the variance within a group was too big compared to the variance between the groups. These parameters were left out because they would not increase the prediction quality. 

---