[ Pobierz całość w formacie PDF ]
in silico log BB models for passive diffusion have assayed (from in-stock compounds), etc.
been included in this review. Tables 2 and 3, Other methods utilized for classification pur-
provide a detailed overview of current progress poses, such as recursive partitioning, are
in this field, from initial steps by Hansch et al. quite powerful not only due to their predic-
to nowadays, as well as the caveats. Following tive power but also as guidelines for medic-
analysis highlights some important points related inal chemists.
to limitations and warnings, application scope and Local models versus General models:
reliability for up to date reported log BB models:
(i) Knowledge about the chemical space
Data sets: As for any in silico model, data sets for the training set is very important
must be as uniform as possible. Ideally, all information as this will help to define
data should be generated from identical the application scope of the model. Other-
experimental conditions and come from the wise, estimations will be based on extra-
same laboratory. If not, the risk for garbage polations to other chemical spaces and
in, garbage out in the models increases, therefore dramatically decrease the per-
resulting in estimations with high uncer- formance of the model.
tainty. In addition, knowing the experimen- (ii) Building up specific log BB models for
tal error in the data set is quite useful since each independent chemical series results
this may help to identify overfitting models in an improvement of their performance
as well as to know where the expectations for and reliability.
the model should be placed.
Descriptors: Rational selection of descriptors Quantitative and Qualitative models: As
involved in passive diffusion, for example, Tables 2 and 3 describe, classification models
those considered in rules-of-thumb for together with those quantitative models
log BB, as well as considering these 5 Abra- utilized as a categorization tool, provide
ham descriptors (LFER), which are highly in general quite good performances with
correlated with the previous, provides models overall accuracies above 70%. Therefore,
with a very good performance. This is a very these are pretty useful prioritization tools
important point for models interpretability in the drug discovery process. But, one key
and utility: not just as prediction tools but factor that should be properly defined to get
also as guidelines in the design of new com- optimal performance for quantitative log BB
pounds. On the other hand, there are algo- estimations as classification tools, is the
rithms to select the most relevant threshold separating BBB penetrating from
DOI 10.1002/jps JOURNAL OF PHARMACEUTICAL SCIENCES, VOL. 98, NO. 12, DECEMBER 2009
4456 MENSCH ET AL.
nonpenetrating. ACKNOWLEDGMENTS
Application Scope for quantitative and
qualitative models within the Drug Discov- The authors would like to acknowledge Dr. Roy
ery process. De Maesschalck from Johnson & Johnson Phar-
(i) Compound prioritization should be maceutical Research & Development, a division of
applied in different phases of the drug Janssen Pharmaceutica N.V., Beerse, Belgium
discovery process: from compounds to be for proofreading and his scientific input to this
synthesized to compounds to be assayed. review.
In this scenario, qualitative linear models
and quantitative models together with
REFERENCES
nonlinear statistical methods ( black
boxes ) fit pretty well.
1. Miller G. 2000. Breaking down barriers. Science
(ii) Rational design. In this case a qualitative
297:1116 1118.
linear model and some quantitative mod-
2. Reese TS, Karnovsky MJ. 1967. Fine structural
el(s) such as recursive partitioning, are
localization of a blood-brain barrier to exogenous
mandatory to identify the key descriptors.
peroxidase. J Cell Biol 34:207 217.
This knowledge is very useful to select
3. Brightman MW, Reese TS. 1969. Junctions
compounds for screening in order to refine
between intimately apposed cell membranes in
the model, as well as to design novel che-
the vertebrate brain. J Cell Biol 40:648 677.
mical structures and focused analogues
4. Minn A, Ghersi-Egea JF, Perrin R, Leininger B,
based on optimal values for the descrip- Siest G. 1991. Drug metabolizing enzymes in the
brain and cerebral microvessels. Brain Res Brain
tors.
Res Rev 16:65 82.
5. Brownlees J, Williams CH. 1993. Peptidases, pep-
tides, and the mammalian blood-brain barrier.
With the increasing knowledge of the numerous
J Neurochem 60:793 803.
active transport processes involved in the blood
6. Brownson EA, Abbruscato TJ, Gillespie TJ, Hruby
brain barrier, it is becoming evident that models
VJ, Davis TP. 1994. Effect of peptidases at the
based on passive diffusion alone, reported in
blood brain barrier on the permeability of enke-
Tables 2 and 3, are not suitable to cover most of
phalin. J Pharmacol Exp Ther 270:675 680.
the options. Therefore, a sequential screening
7. Cordon-Cardo C, O Brien JP, Casals D, Rittman-
approach based on reliable models for passive
Grauer L, Biedler JL, Melamed MR, Bertino JR.
diffusion in combination with models predicting 1989. Multidrug-resistance gene (P-glycoprotein)
is expressed by endothelial cells at blood-brain
P-gp substrate properties may be a good strategy
barrier sites. Proc Natl Acad Sci 86:695 698.
to cover a broad range of drug entry and efflux
8. Krämer SD, Abbott NJ, Begley DJ. 2001. Biologi-
mechanisms.
cal models to study blood-brain barrier permea-
tion. In: Testa B, van de Waterbeemd H, Folkers G,
Guy R, editors. Pharmacokinetic optimization in
CONCLUSIONS
drug research: Biological, physicochemical and
computational strategies. Wiley-VCH: Weinheim.
Various In Vivo, In Vitro, and In Silico methods,
pp. 127 153.
for estimating small molecule transfer across the
9. Deguchi Y, Nozawa K, Yamada S, Yokoyama Y,
[ Pobierz całość w formacie PDF ]
