Using artificial intelligence to mine libraries of untested chemical compounds can speed up the search for potentially useful new drugs. A technique developed by two pharmaceutical chemists promises to evaluate compounds much faster and at less cost than current methods.

Vast repositories of untested compounds reside in compound libraries, but the amount of time and effort necessary to pinpoint those that could be medically useful is prohibitive. These libraries are more akin to drawers full of unlabeled photographs than a true library, making searches haphazard.

Mortar and pestle.Steven Altschuler and Lani Wu of the University of California-San Francisco applied experience working with image recognition software to the problem of locating candidate compounds. Their search method involved creating a new kind of cell, writing software, and sorting through the data that the search retrieved.

Scientists searching compound libraries typically develop a “reporter cell,” designed to change in a unique way when a compound has the desired effect. The researchers’ new type of reporter cell, called ORACL (Optimal Reporter cell lines for Annotating Compound Libraries), can screen for multiple characteristics simultaneously in a process similar to using facial recognition software to sort through a pile of photographs. A recent experiment screened 11,000 drugs from multiple compound libraries for six disease pathways.

To understand all of the information contained in a reporter cell, the research team developed software that can rapidly catalog compound characteristics and compare them to reference drugs. Initial experiments identified 100 new compounds that fit one of six drug classes. In addition, the software identified a number of other potential useful compounds that were not in the experimental design.

The power of ORACL and the companion software in the search for new drugs is potentially game-changing. Researchers will be able describe a type of compound and then search compound libraries for analogues. The speed with which researchers can weed out uninteresting candidate drugs and test promising ones could bring new drugs to market much faster.