however. “Before
we synthesize drug
candidates, we use computers to, for in-
stance, decide which of the astronomical
number of potential compounds would be
the most promising ones,” explains Pro-
fessor Alexander Hillisch, who leads the
Computational Chemistry Department
at Bayer HealthCare in Wuppertal. “For
example, the charge of the virtual mol-
ecules plays an important role and that
we can calculate.” The charge influenc-
es how a substance dissolves, how well
it penetrates membranes and whether
it could cause side effects. To be able
to predict the charge state, Hillisch and
his colleagues examine the functional
groups of the molecule closely. These
components determine the characteristic
chemical properties of the compound.
“The crucial factor that we use to deter-
mine the charge status is what is termed
the pKa value of these groups,” explains
Dr. Mario Lobell, a chemist and software
developer in Computational Chemistry at
Bayer HealthCare.
The pKa value of a compound can al-
so be determined experimentally, but for
the millions of potential drug candidates
and especially for virtual molecules which
have not yet been synthesized, that is
not possible. “Using our calculations, we
can to a certain extent establish which
molecules are promising right on the
computer. We can then synthesize them
in the laboratory and carry on with our
work,” explains Hillisch. Bayer’s scientists
use special software to calculate the pKa
values, but at first it caused problems.
“The programs were relatively slow, they
weren’t very user-friendly and they were
also relatively imprecise,” says Lobell. The
reason was a dilemma faced by the soft-
ware manufacturers, who can only devel-
op and train their programs using publicly
accessible molecular compounds and pKa
data. “But these substances are very dif-
ferent in terms of their structure to our
typical pharmaceutical compounds,” says
Lobell.
As a result, the predictions for medicinal
drug candidates were automatically im-
precise. “It was as if you had crammed
French vocabulary and then had to write
an exam in German,” explains the Bayer
scientist.
To improve the data situation, Bayer’s
experts cooperated with the Califor-
nia-based software company Simula-
Mos doles molupturio denihil-
lore comnis sit omni
The charge of a molecule
determines its action
1
hour
is enough time for the new
software to process more than
100,000 compounds.
Source: Bayer HealthCare
Virtual chemistry for real-life treatments: Dr. Mario Lobell and Professor Alexander Hillisch (photo left, left to right) have co-developed software that
supports optimization of new drug products, such as drug candidates for the treatment of pulmonary hypertension (photo right).
40
Bayer research 28 July 2015
tions Plus Inc. Together they developed
a complex program that determines the
pKa values of molecules significantly
more precisely and faster: it can calculate
more than 100,000 compounds per hour
in unprecedented quality. “That means it
is capable of calculating Bayer Health-
Care’s entire compound library in just
two days,” says Hillisch. “It significantly
accelerates and simplifies drug discovery
and molecule design,” adds Lobell, who
managed the liaison with Simulations
Plus.
The new software uses the chemical
structural formula of a drug candidate
and presents the information in graph-
ic form with the relevant data listed in
tables. It also takes into account import-
ant factors that previous programs could
not factor in, such as the interactions
between several functional groups. For
this, the software has to be trained with
as many known molecules and param-
eters as possible. Simulations Plus had
already built pKa models using data for
some 11,000 compounds known to them
from the open scientific literature. Bayer’s
scientists contributed roughly 19,500 ad-
ditional measured pKa values for some
16,000 pharmaceutical compounds, along
with their molecular structures. “We were
very impressed by the quantity and qual-
ity of data that Bayer provided us with.
And they also gave us a huge amount of
support during the development of the
new model,” relates Dr. Robert D. Clark,
Director of Cheminformatics at Simula-
tions Plus. Together, the experts in Ger-
many and California refined the model
and closely examined any unusual values,
thus making sure that only high-quality
data were used. “Neither Simulations Plus
nor we could have solved these problems
alone,” says Lobell.
Thanks to the innovative predictions
methodology of the software company
in combination with the comprehensive
experimental database and know-how of
Bayer HealthCare, a fundamental scientif-
ic problem has been solved. “The software
is now commercially available worldwide.
In this way we’re fostering progress in
drug discovery and not just keeping this
approach for ourselves,” says Hillisch. And
his team is already looking at new objec-
tives, such as predicting the lipophilicity
or water solubility of drug-like molecules
with a high level of accuracy.
Using mathematical models and vir-
tual patients, the researchers in Bayer’s
laboratories have an increasingly deep
understanding of the routes that drug
products take through the human body
and are thus able to optimize their drug
candidates for patients.
What advantages does predictive software offer drug developers?
Imagine you’re standing in a completely dark room and want to
find the exit. You’ll find it eventually just by wandering around but
it would probably take you a long time. Wouldn’t it be great to have
a flashlight in this situation? Like a flashlight, predictive software
helps us to orientate ourselves when we are searching for answers
to a whole host of questions that are vital to pharmaceutical sci-
ence. It saves money and, even more importantly, time.
What impact will new software and mathematical models have on
medicine?
The predictions derived from these models play an important role in
helping us understand what happens to the active ingredient of a
pill once it has been taken by an animal or human. Computer simu-
lations of basic physiological processes enable us to reliably estimate
dosages for initial trials in humans and to predict how effective a
specific dose is likely to be.
We are also seeing promising approaches for predictions of the reac-
tions in children, pregnant women and the elderly. That reduces the
number of subjects needed to carry out appropriate clinical trials for
these vulnerable populations.
“ Prediction software is
like a flashlight”
Robert
Fraczkiewicz
research
talked to Dr. Robert Fraczkiewicz, lead scientist at the software
developer Simulations Plus Inc., about the collaboration with Bayer.
Software trained with known
pharmaceutical compounds
www.research.bayer.com/
virtual-liver
More information on this subject
Computer models
DOSSIER
Bayer research 28 July 2015
41