predicting the side effects of a medication is not only useful after the drugs are on
the market. catching potential problems before the compounds even reach clinical
trials can save pharmaceutical companies time and money by screening out failing
candidates early on.
researchers at the novartis institutes for Biomedical research and the University of
california, san francisco, applied a method called the similarity ensemble approach
(seA) to the problem. drugs work by binding to and interacting with target proteins
in the body, but they also often bind to off-target proteins and cause side effects. seA
examined the chemical properties of 656 approved drugs to see how similar they were
to other molecules known to bind to 73 different proteins that are associated with side
effects. this was the first time such an informatics approach was applied systematically
to the search for side effects, according to the researchers.
the computer compared the chemical structure of those drugs to a database of more
than 285,000 molecules known to interact with 1,500 human proteins. it predicted
about 1,200 potential interactions, nearly 900 of which had never been explored. the
researchers searched other databases, ran their own chemical tests, and confirmed
almost half of the computer’s predictions. eugen lounkine, a researcher at novartis
and first author of the study, which was published in Nature, says the computer acts as
a screening tool to make the development of medication more efficient. “it’s early in
drug discovery, where you have more compounds than you have resources to test with
biochemical assays,” he explains.
lounkine says finding off-target proteins can also help explain some previously
unexplained side effects. for instance, chlorotrianisene, a synthetic estrogen sometimes
used to treat prostate cancer, can cause upper abdominal pain, but no one knew why.
the novartis study discovered the drug interfered with an enzyme known as cOX- 1 in
much the same way a blood thinner, which is known to cause such pain, would.
Informatics, Drugs,
and Chemical Properties
one correspondence between those is
really daunting.”
The reports are “spontaneous,”
meaning they are not standardized
and are based on individual judgments
about symptoms someone noticed and
deemed significant. The data is sta-
tistically noisy, full of biases and con-
founding factors that may not be eas-
ily identifiable. One well-known bias,
for instance, is what Tatonetti calls
“the Vioxx effect”; when a link was dis-
covered between the painkiller Vioxx
and heart attacks, the resulting pub-
licity prompted people using Vioxx to
report more heart-related symptoms,
which made the background rate for
those symptoms seem greater than
normal, thereby masking the drug’s
real effects. There are also symptoms
that might be associated with a drug
but are not caused by it. Someone tak-
ing a medication for diabetes, for in-
stance, could have symptoms caused
by the underlying disease, though an
algorithm would only notice the asso-
ciation between the symptoms and the
drug, and could incorrectly conclude
the drug was causing the problem.
Someone using an arthritis medicine
might report complications that are
the result of being elderly, and not
from the medication. Modern signal-
detection algorithms try to account for
biases, but have not addressed all the
possible sources, the researchers say.
“There are a lot of scientific compu-
tational challenges to this database,”
says Robert O’Neill, director of the Of-
fice of Biostatistics in the FDA’s Center
for Drug Evaluation and Research.