Virtual Screening Lab Zeroes in on New Drugs

Researchers at Rensselaer Polytechnic Institute (RPI) have come up with computational tools that serve as a virtual screening lab to help chemists weed through millions of possible drug candidates even before they dirty their first test tube.

Chemist Curt Breneman, mathematician Kristin Bennett, and computer scientist Mark Embrechts developed faster and more accurate techniques for describing molecules and combined them with next-generation neural networks and learning methods as part of the Drug Discovery and Semi-Supervised Learning (DDASSL) project.

Funded by a $1.2 million National Science Foundation Knowledge and Distributed Intelligence award, the DDASSL (pronounced “dazzle”) project has spawned a number of descendants. Today, 10 research projects on the RPI campus, ranging from the life sciences to materials science to cybersecurity, can trace their origins in part to DDASSL (http://www.drugmining.com/).

In addition, Concurrent Pharmaceuticals, based in suburban Philadelphia, is evaluating the DDASSL techniques in a real-world environment. According to Jean-Pierre Wery, Concurrent’s vice president of computational drug discovery, the information used by DDASSL is different from what has been used traditionally. “These tools are consistent with Concurrent’s efforts to change the way we think about the drug discovery process,” Wery said.

When starting to develop a new drug to attack a particular biological target, a pharmaceutical chemist is confronted with the accumulated knowledge stored in vast public and corporate databases on tens of millions of potential drug molecules and their effects.

DDASSL techniques and a relatively inexpensive Linux cluster provide a fast and accurate tool for pinpointing the likeliest candidates from these databases. DDASSL can screen 10 million molecules per day for their potential drug interaction with a model of the biological target molecule.

By comparison, the best virtual screening techniques prior to DDASSL would use less accurate molecule descriptors and still examine less than a million molecules in a day. Actual laboratory tests top out at several hundred or thousand per day even with the latest high-throughput equipment.

“DDASSL tools are also an easy way to take an idea and run it past a model,” Breneman said. “Chemists can test their ’wild ideas’ quickly and without the expense of a lab test.”

NSF Program Officer: Maria Zemankova, (703) 292-8918, mzemanko@nsf.gov

Principal Investigators:
Curt Breneman, brenec@rpi.edu, (518) 276-2678
Kristin Bennet, bennek@rpi.edu, (518) 276-6899
Mark Embrechts, embrem@rpi.edu, (518) 276-4009