Outlier ratings

Second, the descriptors for a compound should fit the pattern of interdescriptor correlations implicit in the PLSR model. The compounds with a substantially different descriptor values are detected using the outlier rating that is based on a relative residual descriptor variance for a prediction object i compared to the residual training variance ... 

After predicting the activity and selectivity values by means of the MFTA models, the compounds with unreliable predictions were filtered from the virtual screening set using the outlier rating values (Section 5.2.2.6) for two predictions (the threshold value was 30). As a result, a set of 50 structures was obtained. Figure 5.8 shows the distribution of the predicted values. Only one compound is predicted to provide better selectivity than the training set structures as well as reasonable activity. The local descriptor values for this adenosine derivative indeed match the descriptor impacts identified by the MFTA-based selectivity model. 

Figure 5.15 Distribution of the predicted endpoint and outlier rating values for the activity (a) and selectivity (b) of the AMPA receptor antagonists.

Cropley made general recommendations to develop kinetic models for compUcated rate expressions. His approach includes first formulating a hyperbolic non-linear model in dimensionless form by linear statistical methods. This way, essential terms are identified and others are rejected, to reduce the number of unknown parameters. Only toward the end when model is reduced to the essential parts is non-linear estimation of parameters involved. His ten steps are summarized below. Their basis is a set of rate data measured in a recycle reactor using a sixteen experiment fractional factorial experimental design at two levels in five variables, with additional three repeated centerpoints. To these are added two outlier... 

For the reasons described, no specific test will be advanced here as being superior, but Huber s model and the classical one for z = 2 and z = 3 are incorporated into program HUBER the authors are of the opinion that the best recourse is to openly declare all values and do the analysis twice, once with the presumed outliers included, and once excluded from the statistical analysis in the latter case the excluded points should nonetheless be included in tables (in parentheses) and in graphs (different symbol). Outliers should not be labeled as such solely on the basis of a fixed (statistical) rule the decision should primarily reflect scientific experience. The justification must be explicitly stated in any report cf. Sections 4.18 and 4.19. If the circumstances demand that a mle be set down, it is best to use a robust model such as Huber s its sensitivity for the problem at hand, and the typical rate for false positives, should be investigated by, for example, a Monte... 

Note on GMPs The assays are conducted on individual dosage units (here tablets) and not on composite samples. The CU test serves to limit the variability from one dosage unit to the next (the Dissolution Rate test is the other test that is commonly used). Under this premise, outlier tests would be scientific nonsense, because precisely these outliers contain information on the width of the distribution that one is looking for. The U.S. vs. Barr Laboratories Decision makes it illegal to apply outlier tests in connection with CU and DR tests. This does not mean that the distribution and seemingly or truly atypical results should not be carefully investigated in order to improve the production process. 

The o scale was developed to incorporate the effect of through-resonance. Both types of correlations included ortho- as well as meta- and para-substituted phenoxides, and the only outliers are compounds that exhibit strong intramolecular hydrogen bonding because this effect is not incorporated in most molecular descriptors. The rate constants for oxidation of phenoxide anions give good Hammett correlations to cr constants, as shown in the following equation ... 

Four batches of the same sort of composite rocket propellant have been cast in a lab. After curing, the experimental rocket motors were static fired at normal temperature. The following burning-rate values at P=70bar pressure have been obtained 14.199 14.531 14.197 14.193. Does the burning-rate value V=14.531mm/s represent the outlier with 95% confidence level ... 

FIGURE 6 Summary of measurements of in situ rates of bacterial DOC consumption, calculated as the sum of bacterial production and respiration, and in vitro rates of DOC consumption, calculated from declines in DOC during batch incubations. Box-and-whisker plots show median, and upper/lower quartiles (box), and the range of values (bars). Extreme outliers are denoted by open circles. The in situ consumption rates are from the global dataset of simultaneous measurements of bacterial production and respiration collected by del Giorgio and Cole (1998), with the addition of unpublished data (see text). Because the in situ rates have been measured under a range of temperature, for this comparison the bioassay rates have not been corrected for temperature. 

The full model can reproduce all the systematic features of the absorption curves in Fig. 2.5 [11]. In the present form, the model does not account for the interpatient variability observed in Fig. 2.4 or for the nonsystemahc phenomena discussed in connection with Fig. 2.5. Ideally, in our mechanism-based modeling approach, any form of unsystematic variation in the experimental results should be given a separate explanation, i.e. specific studies should be undertaken to explain the observed variability in absorption rates and binding capacities. Statistical outliers should be considered as potential sources of new information. 

Pressure drop. The pressure drop across 24 freshly prepared tubes was determined at a flow rate of 300 ml/minute. The range was 25.0 - 44.0 mm Hg. The mean was 36.7 5.4 (la ) mm Hg. One outlier was excluded. Conversely, the variation of the flow rate was determined at a constant pump setting, such that the flow rate of one tube used as a reference was 300 ml/minute. The pressure drop was 20.5 mm Hg, the range was 275 - 315 ml/minute. The mean was 292 11 (la ) ml/minute. Thus, the coefficient of variation was 0.037. 

Structure-based clustering is used to group related compounds for the purpose of HTS data analysis, identification of SAR series, and detection of potential outliers (Engels et al., 2002). In one example, researchers at GNF reported a statistical approach to dynamically score each scaffold family (obtained by prior clustering of screened structures) based on family members HTS activities. This method identifies compounds that share structural similarities and similarly high HTS activities it yielded greatly improved confirmation rates compared to using a static (scaffold-independent) activity cut-off (Yan et al., 2005). 

In practice, relaxation parameters are measured as a series of HSQC-type spectra as a function of a variable time delay, and peak heights (or volumes) are used to fit relaxation rates.73 The next step is the determination of the overall rotational correlation time of the molecule, typically from the T /T2 ratio obtained for the residues. In such calculations, outliers are generally excluded, that is those residues for which Rex or ie is significant (residues with T, /T2 ratio well above and below the average, respectively, Ref. 71). The second step is to determine the best fit between the relaxation parameters Ri, R2 and heteronuclear NOE and the motional parameter sets. The correspondence between these can be expressed... 

As part of the evaluation of results of the intercomparison exercises for major and trace elements, NRC assigns a performance evaluation criteria based on the number of times results reported by a laboratory fall within acceptable criteria. The percentage of laboratories achieving superior or good performances has increased since 1991 from 46% to 83% (Figure 13.3). Superior-rated laboratories submitted results for most analytes within the 95% confidence intervals goodrated laboratories submitted many results within the accepted range with a minimum number of outliers (16). 

IQR=Q3-Q , where Q3 is the third quartile and Qj is the first quar-tile, when the median divides the experimental sample into two parts. Outliers were found in experiments No 1, 3 and 6 and may be rejected but we did not find enough good reason for that and they were used in calculations of the average corrosion rate value and its standard deviation. 

There is rather significant scattering of data and some of the eight experiments may be regarded as outliers. Therefore we have tested the null hypothesis H0 of equality of the lowest mean of corrosion rate in experiment No 1 and the highest mean in experiment No 8. The calculated value F=s12/s82=5.92/5.02=1.4 for standard deviations was compared with Fisher distribution statistical test values... 

Since the distribution of the log-transformed data after removing these outliers satisfied the Shapiro Wilk normality test (Fig. 8. A2b), the cut point determination for these validation data was made using the parametric method (i.e., mean + 1.645 x SD). The threshold 1.645 for normally distributed data ensures 5% false-positive rate. This cut point of the log-transformed values of the sample replicate means was determined for each analyst, each assay run, and for all the data combined (Table 8.A2). 

The direct Unear plot of Eisenthal and Comish-Bowden is a more recent introduction. It is not an attractive method from the point of view of display since each measurement is represented as a line rather than a point (Fig. 2d). Its claim, however, is to statistical superiority for the purposes of (non-manual) computation. The estimation of kinetic parameters by this method is claimed to be less sensitive to undue distortion by outliers - i.e. a few exceptionally inaccurate rate measurements. 

Figure 8 °Be/ Be ratios in Pacific (A, B) and Atlantic (C, D) Fe-Mn crusts (Ling et al., 1997 von Blanckenburg and O Nions, 1999). The smooth exponential decrease with depth usually observed in the measured data (as fitted following the stippled lines in A, C, where some individual outliers are attributed to alteration) is compatible with radioactive decay of Be only (Ti/2 = 1.5 My). A kink in the fits to the data is due to changes in growth rates. Changes in the initial ratio would be visible in the form of offsets in the data. At the time resolution of the samples taken (several hundred thousand years) these are not observed. Therefore, the derived growth rates can be used to time-correct the °Be/ Be ratios for radioactive decay and to calculate initial ratios. The results shown in B and D indicate that the °Be/ Be ratios in both the Pacific and the Atlantic have been within the range of modern sea water, for the last 7-10 My, despite presumably widely varying erosional input into the ocean.

---