Blood Brain Partitioning

Whether drug design criteria require significant brain penetration or demand minimal partitioning into brain tissue, estimation of the brain-blood ratio, BB = [brain]/[blood], is essential in drug design. In our first study, we organized experimental data, given as the logarithm, log BB, from 11 literature sources to create a very diverse set of 106 compounds.28 MLR methods led to models, but the linear technique yielded poor statistics. When nonlinear terms were developed for specific structure descriptors, an adequate model was obtained: r2 = 0.66 and s = 0.45. Two issues arose that appear typical of ADMET modeling. First, it was necessary to introduce nonlinear terms into the model in order to achieve useable results. Second, four compounds appeared to be outliers and were removed. Although the exact reasons for outlier status are not clear, two explanations were considered: (1) the eclectic nature of the experimental data, which include several types of experimental protocols, and (2) the possibility that salient structure features are not adequately encoded in the descriptor set. Because this data set is eclectic from the experimental protocol point of view, no reliable estimate of the actual experimental error for the data set is available. However, statistics from cross-validation indicate the model to be useful for drug design purposes.

This model for blood-brain barrier partitioning was also directly interpretable in terms of molecular structure, leading to statements about the manner in which calculated blood-brain barrier partitioning varies with molecular structure as a guide to molecular modification. For example, one of the descriptors in the model is the hydrogen E-state value for hydrogen bond donors (see 4.22 Topological Quantitative Structure-Activity Relationship Applications: Structure Information Representation in Drug Discovery). The model indicates that as the descriptor for hydrogen bond donor strength increases, log BB decreases, in agreement with the assumed nature of membrane passage across the blood-brain barrier. Furthermore, the model indicates that the contribution of an hydrogen bond donor group covers a range of values, depending on the bonding environment of the donor group. The range in contribution is traced directly to the fact the E-state value for an atom or group encodes its molecular bonding environment. This bonding environment aspect of E-state descriptors highlights their significant advantage over a simple fragment count, which is insensitive to molecular environment.

In one additional study with the model for blood-brain barrier partitioning, log BB values were predicted for a set of 20 039 organic molecules from the Pomona MedChem database.28 The experimental log BB values were not available for these compounds, but the range of predicted values could indicate the potential for realistic predictions. In this test, 99.3% of the predicted values fell within the same log BB range as the training set. No unreasonable values were computed by the model.

Subsequent unpublished work has extended the range of applicability of the BB model. To deal with the nonlinear nature of the property-structure relation, an ANN model was developed for 103 compounds, and tested by the prediction of an external validation test set.33 Figure 7 shows the predictions for this external test set. One significant aspect of the new validation test set is that it consists of data on human brain and human cerebrospinal fluid (CSF) as well as data on rat and monkey. Data in the training set are entirely based on rat brain, but the external validation test set includes human brain data as well as CSF data in addition to monkey brain. The validation test results are statistically satisfactory, rpress = 0.62 and MAE = 0.39, even though the experimental nature of the test data arises from different sources than the training set. To provide a different view of the predictions, the data were divided into three bins, corresponding to low, medium, and high partitioning. These results are presented in Figure 8. Note that 83% of the predictions fall into the correct bin. Furthermore, no prediction on a compound in the high bin falls into the low bin, and vice versa.

Figure 7 External validation test set predictions on 74 commercial drugs for blood-brain barrier partitioning, showing log BB predicted by the model versus the experimental value. The model is based entirely on data from rat brain. This validation set consists of data from several types of measurements: 18, rat brain; 44, human CSF; 5, human brain; 2, monkey brain; 3, other animals. Eighty-eight percent of the predictions are within 0.75 log units of the experimental values.

Experimental log BB

Figure 7 External validation test set predictions on 74 commercial drugs for blood-brain barrier partitioning, showing log BB predicted by the model versus the experimental value. The model is based entirely on data from rat brain. This validation set consists of data from several types of measurements: 18, rat brain; 44, human CSF; 5, human brain; 2, monkey brain; 3, other animals. Eighty-eight percent of the predictions are within 0.75 log units of the experimental values.

Figure 8 External validation test set predictions on 74 commercial drugs for blood-brain barrier partitioning as log BB given in three bins, indicating that the predictions of the model yield 83% correct placement in the high, medium, and low categories. The model is based entirely on data from rat brain. This validation set consists of data from several types of measurements: 18, human brain; 44, human CSF; 5, human brain; 2, monkey brain; 3, other animals.

"D

Medium

High

Predictions in experimental bin Predictions off ±1 bin Predictions off ±2 bins

Figure 8 External validation test set predictions on 74 commercial drugs for blood-brain barrier partitioning as log BB given in three bins, indicating that the predictions of the model yield 83% correct placement in the high, medium, and low categories. The model is based entirely on data from rat brain. This validation set consists of data from several types of measurements: 18, human brain; 44, human CSF; 5, human brain; 2, monkey brain; 3, other animals.

Lower Your Cholesterol In Just 33 Days

Lower Your Cholesterol In Just 33 Days

Discover secrets, myths, truths, lies and strategies for dealing effectively with cholesterol, now and forever! Uncover techniques, remedies and alternative for lowering your cholesterol quickly and significantly in just ONE MONTH! Find insights into the screenings, meanings and numbers involved in lowering cholesterol and the implications, consideration it has for your lifestyle and future!

Get My Free Ebook


Post a comment