Optimal HTS Fingerprint Definitions by Using a Desirability Function and a Genetic Algorithm

The
use of compound biological fingerprints built on data from
high-throughput screening (HTS) campaigns, or HTS fingerprints, is
a novel cheminformatics method of representing compounds by integrating
chemical and biological activity data that is gaining momentum in
its application to drug discovery, including hit expansion, target
identification, and virtual screening. HTS fingerprints present two
major limitations, noise and missing data, which are intrinsic to
the high-throughput data acquisition technologies and to the assay
availability or assay selection procedure used for their construction.
In this work, we present a methodology to define an optimal set of
HTS fingerprints by using a desirability function that encodes the
principles of maximum biological and chemical space coverage and minimum
redundancy between HTS assays. We used a genetic algorithm to optimize
the desirability function and obtained an optimal fingerprint that
was evaluated for performance in a test set of 33 diverse assays.
Our results show that the optimal HTS fingerprint represents compounds
in chemical biology space using 25% fewer assays. When used for virtual
screening, the optimal HTS fingerprint obtained equivalent performance,
in terms of both area under the curve and enrichment factors, to full
fingerprints for 27 out of 33 test assays, while randomly assembled
fingerpints could achieve equivalent performance in only 23 test assays.