At the early stage of drug discovery, many thousands of chemical compounds
can be synthesized and tested (assayed) for potency (activity) with high th
roughput screening (HTS). With ever-increasing numbers of compounds to be t
ested (now often in the neighborhood of 500,000) it remains a challenge to
find strategies via sequential design that reduce costs while locating clas
ses of active compounds.
Initial screening of a modest number of selected compounds (first-stage) is
used to construct a structure-activity relationship (SAR). Based on this m
odel, a second-stage sample is selected, the SAR updated and, if no more sa
mpling is done, the activities of not yet tested compounds are predicted. I
nstead of stopping, the SAR could be used to determine another stage of sam
pling after which the SAR is updated and the process repeated.
We use existing data on the potency and chemical structure of 70,223 compou
nds to investigate various sequential testing schemes. Evidence on two assa
ys supports the conclusion that a rather small number of samples selected a
ccording to the proposed scheme can more than triple the rate at which acti
ve compounds are identified and also produce SARs effective for identifying
chemical structure. A different set of 52,883 compounds is used to confirm
our findings.
One surprising conclusion of the study is that the design of the initial sa
mple stage may be unimportant: random selection or systematic methods based
on chemical structures are equally effective.