Taking a “Chem-Informatics” Approach

To improve predictability, Breneman, professor of chemistry at Rensselaer, works in the area of “virtual screening.” To develop a new drug costs nearly $1.7 billion, according to industry analysts. The tremendous winnowing-out process of drug screening is a major factor in the expense. Simply put, the farther along a compound gets, the more money it costs. Applying computational methods as part of an early screening process can be a huge benefit to drug makers.

Breneman leads a team of Rensselaer researchers with expertise in computer science, chemistry, and mathematics who have collaborated to create a software package capable of quickly identifying molecules that show promise for future medicines. This combination of programs package, now being licensed to companies, enables drug makers to comb through enormous databases of molecules and identify the ones that have sound medicinal properties. Of equal importance, the virtual screening process helps to identify compounds that may cause serious side effects when they are still early in the discovery process.

The software is part of the Drug Discovery and Semi-Supervised Learning Project (DDASSL, pronounced “dazzle”), that was supported by a $1.2 million grant from the National Science Foundation. (To read more, go to “Predicting New Medicines”)

Breneman’s work focuses on two major areas of drug activity: toxicity and pharmacokinetics. Toxicity of course refers to unwanted and often harmful side effects of drugs. Pharmacokinetics refers to the processing of drugs by the body, comprising absorption, distribution, metabolism and excretion.

Toxicity can be assessed with models for specific side effects that can doom future success of a drug. One example with which Breneman has worked is called hERG channel inhibition. During normal heart function, the heart muscle contracts in rhythmic response to the action of a natural pacemaker system involving ion channels, such as hERG, that gate on an off at specific times. “If you inhibit the hERG channel, your heart beat becomes unusual and elongated,” says Breneman. “One beat runs into another beat, and you can get a fatal arrhythmia.” He cites a class of antihistamines that were pulled from the marked because of this side effect. “Nobody knew about it then. Now they want to filter potential drugs like this out computationally before they even get into the pipeline.”

Breneman’s group creates programs that analyze a molecule’s 3-dimensional structure and describe it numerically. Some of the numerical descriptors are obvious, says Breneman, like size, volume, and shape, while others are less so, like how many rings or branches the molecule has.

But structural descriptors only go so far, says Breneman. To go further, he models the electronic properties of a molecule. Using programs developed by his research team, “we compute the properties of the electron density fields around the molecule and reduce those to a fairly large set of numbers.” This numerical representation works as a whole to describe the properties of a molecule. “You can’t really put an interpretation on any individual one,” he says.

The premise of this approach harks back to quantum mechanics, says Breneman. “If you knew something about the distribution of electrons, you could really tell just about everything that molecule was capable of doing,” he says. “Our job in this chem-informatics approach is to find ways of extracting that information. It’s like squeezing blood out of a stone. You’re trying to find ways of probing molecular electron density distributions to get lots of fundamental data.”

Applying his models to drug discovery, Breneman describes how the information can be used. “You can do a go—no-go sorting,” he says. For hERG channel inhibition, his model has proved to be 90 percent correct on molecules that inhibit and 88 percent correct on those that do not.

 “What that means is you can enrich the molecules in your pool that are not going to inhibit hERG.” Because the model is not perfect, some toxic compounds will slip through and the hERG toxicity won’t be caught until later in the drug screening process. The advantage of the computational modeling screen is that far fewer drugs will be fed into the slower and costlier laboratory screening.

“It’s an assistance, a streamlining approach,” Breneman says, “to speed up the process and aid the medicinal chemists in making good decisions.” A good decision at that early juncture has been estimated to make a $100 million difference in the cost of a drug, he says.

Related Web sites:

Rensselaer Research

Drug Discovery and Semi-Supervised Learning Project

Department of Chemistry and Chemical Biology

Department of Chemical and Biological Engineering

Department of Decision Sciences & Engineering Systems

Additional Information:

Predicting New Medicines

New Software Developed at Rensselaer Predicts Promising Ingredients for New Drugs