A Screening Pattern Recognition Method Finds New and Divergent Targets for Drugs and Natural Products

Computational target prediction methods
using chemical descriptors
have been applied exhaustively in drug discovery to elucidate the
mechanisms-of-action (MOAs) of small molecules. To predict truly novel
and unexpected small molecule–target interactions, compounds
must be compared by means other than their chemical structure alone.
Here we investigated predictions made by a method, HTS fingerprints
(HTSFPs), that matches patterns of activities in experimental screens.
Over 1,400 drugs and 1,300 natural products (NPs) were screened in
more than 200 diverse assays, creating encodable activity patterns.
The comparison of these activity patterns to an MOA-annotated reference
panel led to the prediction of 5,281 and 2,798 previously unknown
targets for the NP and drug sets, respectively. Intriguingly, there
was limited overlap among the targets predicted; the drugs were more
biased toward membrane receptors and the NPs toward soluble enzymes,
consistent with the idea that they represent unexplored pharmacologies.
Importantly, HTSFPs inferred targets that were beyond the prediction
capabilities of standard chemical descriptors, especially for NPs
but also for the more explored drug set. Of 65 drug–target
predictions that we tested in vitro, 48 (73.8%) were
confirmed with AC50 values ranging from 38 nM to 29 μM.
Among these interactions was the inhibition of cyclooxygenases 1 and
2 by the HIV protease inhibitor Tipranavir. These newly discovered
targets that are phylogenetically and phylochemically distant to the
primary target provide an explanation for spontaneous bleeding events
observed for patients treated with this drug, a physiological effect
that was previously difficult to reconcile with the drug’s
known MOA.