PREDICT score

Tab. 5.3 Correlation coefficients between the similarity-based and hypothesis-based activity prediction scores and experimental activity values, taken over the whole Cox2 inhibitor set and with...

A reliable druglikeness predictor would give high prediction scores only to compounds that have satisfactory properties based on aU of these criteria, or in other words, there would be few false positives. There has been considerable effort expended over the last 10 to 20 years in modeling individual components of this process, including solubility [18-36], ADME properties [53-70], and toxicities [71-88]. Individually, each of tliese predictions has false positives and false negatives, so it is difficult to expect... 

Fig. 15.18 Selected examples of the distribution of predicted PSA scores for diversity reagents within a 2400-member library built around a common core. Topological PSA scores were calculated according to the method published by ErtI and expressed in A. Significant number of predicted scores greater than 140 is sufficient criteria for deselection of that diversity reagent.

Fig. 15.19 Exampies of the distribution of predicted hERG binding iikeiihood for diversity reagents within a 2400-member iibrary buiit around a common core. Binding to the hERG channei was predicted according to the method reported by Roche scientists [77]. Predicted scores of iess than 0.3 are flagged as potential for concern. Scores above 0.7 are not flagged as potential hERG binders. There is no basis for interpretation for scores between 0.7 and 0.3.

The PCA prediction scores and residuals can then be used to generate two common quality metrics for conveying the abnormality of the measured response namely the Hotelling statistic and the Q residual statistic, which are defined below ... 

Robust epidemiologic evidence has identified an inverse relationship between HDL-cholesterol levels and CHD risk. Indeed, HDL-cholesterol is included in the Framingham CHD risk prediction scores (29). HDL-cholesterol protects against... 

Balancing these three rates equally yields an optimal threshold of 0.846. Both thresholds are indicated by vertical lines in Figure 8. Figure 9 shows the actual predicted scores for the molecules that do cross the blood-brain barrier (BBB+) as well as those that do not (BBB-). The false positive and false negative rates are, of course, direct consequences of which threshold is chosen. The appropriate threshold depends on the goal. For example, if the project is a neuroscience project where BBB+ is the goal, it may be that the team wants to find and reject... 

Figure 9. Random forest BBB predicted scores for molecules assigned as BBB+ and BBB — horizontal reference lines correspond to two decision thresholds. All predictions (scores) are for molecules not in the training set.

Obtain the predicted scores t, for this sample, using the loadings in step 6 and the scaled vector x obtained in step 7. 

For each dataset calculate the predicted scores for the first PC given by Ft = FX.Fp and st = sX.sp. Then recalculate the predicted datasets using models (a) and (b) by multiplying the predicted scores by the appropriate loadings, and call these FZ and SZ. 

Remove sample 1 from the dataset, and calculate the five nonzero PCs arising from samples 2-6. What are the loadings Use diese loadings to determine the predicted scores t =x.p for sample 1 using models based on one, two, three, four and five PCs successively, and hence the predictions x for each model. 

Although most applications were of the cherry-picking type design, the combinatorial design of new chemical libraries should also be feasible. In this case, the scores obtained with the various models can be used to sort the virtual library, followed by building block frequency analysis cf. Focus2D) to determine which reagents should be used in chemical synthesis. Alternatively, combinatorial optimization approaches, such as those in described in ref. 4, can be applied where the model-predicted scores are used as the objective function for optimization. 

Selected scores have been proposed for stratifying risk after MI. These scores have been derived either from clinical trials (TIMI, PURSUIT, GUSTO, etc.) or from registries and cohort studies (PREDICT, CCP, etc.). The majority of them divide the ACS into two groups with and without ST-segment elevation (STE-MI or STE-ACS vs NSTE-MI or NSTE-ACS). This classification is very useful for a better approach of treatment. The GUSTO score includes QRS duration and ECG (Hathaway et al., 1998a,b) prior MI (Table 8.4), and the PREDICT score uses other ECG parameters (ECG severity score) that include ST, Q wave and branch block criteria (Jacobs etal., 1999 Table 8.5). 

Table 8.5 PREDICT score components, definitions, and risk computation.

Also, the inclusion of EF and different biomarkers added significant prognostic information over TIMI and PREDICT scores. Recently, other markers, such as CRP, and different interleukins have been added to the risk assessment (Anguera et al, 2002 Zairis et al, 2002), as well as BNP (Bassan et al, 2005) and PAPP (Heeschen et al, 2004). Even the value of multiple biomarkers added to the value of quantitative ST-segment depression has been recently published (Westerhout et al, 2006). 

Ha et al. [150] have shown that DOCK/PMF predicted scores correlate significantly with the measured binding affinities for the highest scoring binding modes of 61 stromelysin/inhibitor complexes [150]. 

The Challenge of Affinity Prediction Scoring Functions for Structure-Based Virtual Screening... 

---