Average fold error

Volume of distribution predicted from tissue composition-based equations showed an average fold error of 2.2 and the correlation of predicted versus observed volume for the five compounds was poor. For the prediction of volume of distribution it was assumed that the volume in human (L/kg) was the same as the observed volume in the rat (ranging from 0.9 to 2.8 L/kg for the five compounds). Due to the uncertainty in the prediction of volume, the error range associated with this parameter was set as a uniform distribution over a twofold range. 

The average fold error (AFE) across this correlation was 1.36. External validation of this approach predicted the human volume of distribution for an additional 9 compounds with an AFE of 1.83. It was recommend that volume of distribution from rat and dog is sufficient to provide a useful prediction of human volume, and that measurement in further species is only warranted if deviations unrelated to plasma protein binding differences are observed in the first two species. Deviations may be caused by active processes which may not scale/translate directly across species. 

Table 16.6 Average fold error and absolute average fold error values for eD2M versus C n normalized clinical dose.

It is worth noticing that as the map units are increased, the average quantization error (qj decrease s since the data samples are distributed more sparsely on the map, whereas the values of the topographic error (l) increase because an unnecessary fold (train) occurs (Polzlbauer 2004). 

The effect could be considerable for solvent-exposed parts of the backbone and could render the NOE values inaccurate. These systematic errors could be minimized by using water flip-back pulses in order to avoid saturation of H20 magnetization [11]. The NOE data are generally more susceptible to errors than Ri and R2 because (i) the NOE experiments start with the equilibrium 15N magnetization that is 10 fold lower than that of (XH) in the Hi and R2 experiments, hence relatively low sensitivity, and (ii) the NOE values are derived from only two sets of measurements, whereas R1 and R2 data are obtained from fitting multiple sets of data the latter is expected to result in a more efficient averaging of experimental errors. 

FIGURE4.38 Ridge regression for the PAC data set. The optimal ridge parameter AK —4.3 is evaluated using repeated 10-fold CV. The resulting (average) predictions (in black) versus the measured y-values are shown with two different scales because of severe prediction errors for two objects. 

Equilibrium properties under each set of conditions were obtained by averaging over 10 trajectories. Each trajectory starts with an equilibration period of 105 time steps, followed by a data collection period of 5 x 105 time steps in the folded state and 5 x 107 time steps at the midpoint and unfolded states. Error bars were assumed to be given by one standard deviation. 

Natural proteins are able to fold to specific structures because, on average, native-like interactions between residues are more stable than non-native ones. The former are therefore more persistent and the polypeptide chain is able to find its lowest energy structure by a process of trial and error. Moreover, if the free energy surface or landscape has the right shape (see Fig. 13.1), only a minute fraction of all possible conformations is sampled by any given... 

Figure 5.7 Linearity of CCA fluorescence enhancement. Two-fold serial dilutions of IRDye 800CW-labelled streptavidin are plotted. Each plotted point represents the average of 4 measurements, with one standard deviation shown by error bars. Note that only the error bars for the sample at 0.5pg is large enough to be visible on the graph.

---