Software That Quickly Identifies Promising Molecules

New software can quickly screen large databases and accurately predict the molecules that show potential for future medicines.

The software was developed by researchers at Rensselaer Polytechnic Institute (Troy, NY, USA) with skills in computer science, chemistry, and math. The work was supported with a U.S.$1.2 million award from the U.S. National Science Foundation, and a team that included Curt Breneman, professor of chemistry; Kristin Bennett, associate professor of mathematics; and Mark Embrechts, associate professor of engineering systems.

"The trick with drug discovery is to have the drug molecule fit like a key in a lock, because shape affects its performance,” explained Dr. Embrechts. The safety and effectiveness of medicines depend on the shape and chemistry of the molecules. To find the most likely molecules, the new software makes use of two shortcuts in chemistry and math that enable the computer to rapidly search a vast molecular database.

The first shortcut describes the molecule, its shape, and its chemistry in terms of numbers a computer can rapidly calculate. "Dr. Breneman has a technique to calculate electronic properties on the surface of a molecule very quickly,” noted Dr. Embrechts. "It produces a description, basically a set of numbers, that the computer can use easily.”

The second shortcut identifies which molecules have the right chemistry for a specific therapy. Using advanced pattern-recognition techniques known as kernel methods, the software analyzes a small sample database to identify molecules with the right chemical features. Once the key features are identified, the software can quickly screen large databases, accurately predicting the molecules that show potential.

"Conventional techniques are not truly predictive and don't work,” said Dr. Bennett. "So we borrowed pattern recognition techniques already used in the pharmaceutical industry and added algorithms based on support vector machines. That gives us techniques to predict which molecules are promising.”

The researchers noted that predictive modeling is one of a new breed of drug discovery methods that marks a shift in industry practice, a shift away from cell-based assays performed in the laboratory toward math-based models calculated on a computer.

"Our program allows researchers to ‘crash test' lots of molecules quickly and inexpensively,” said Dr. Breneman. "That prevents a lot of false starts. The ultimate pay-off of this methodology may be that it can support the rapid invention of new drugs when diseases develop quickly and threaten society.”

As drug developers increasingly target complex, chronic illness, drug development becomes far more costly and time consuming. Meanwhile, in the search for new drugs, 99.9% of compounds tested ultimately fail. Accordingly, drug makers want to be able to predict more accurately which compounds will produce the next blockbuster drug.





Related Links:
Rensselaer Polytechnic