Latest News


In Search of Tomorrow's Cures

Published on April 5, 2010 by Aaron Dubrow

Computer prediction of a novel inhibitor binding to the JNK substrate docking site.

Drugs enter our bodies as small molecules and bind to the surface of target proteins, inhibiting their function or reproduction. For a compound to tame a headache or reduce a swollen knee, it needs to be effective at small doses, and selective enough to limit side effects.

With so many medicines to choose from on the shelves of your local pharmacy, it would seem that creating a new drug is a simple, straightforward process. In reality, discovering a new drug can be a Herculean effort.

According to a prominent paper by Joseph DiMasi of Tufts University, it requires 15 years and more than $800 million in research and development for a drug to come to market. This drives up the price of blockbuster drugs, while limiting research into less profitable medications.

Due to the time and costs involved, computing is crucial to drug discovery efforts. By creating virtual models of proteins and ligands, and automating aspects of the interaction process, chemists have been able to screen the pool of possible compounds and discover drugs more efficiently and at a lower cost. Early examples of such discoveries include important HIV protease inhibitors.

However, computational drug discovery has fallen short of its promise, in part because of the inconsistency of conventional docking algorithms, which are used to narrow potential compounds from millions to hundreds, at which point they can be studied in the lab.

This virtual "enrichment" offered by traditional docking is only helpful if the most effective molecules end up in the top 10 percent of the prediction. Unfortunately, more often than not, they don't. In fact, studies have shown that docking methods systematically miss some of the best compounds, while promoting duds, leading to frustration and skepticism in the field.

Pengyu Ren, assistant professor of biomedical engineering at The University of Texas at Austin, is trying to solve this problem by taking advantage of sophisticated physical models and supercomputing power to create a more robust way of searching for new drugs.

Collaborating with the Texas Advanced Computing Center (TACC) and the Texas Institute for Drug and Diagnostic Development (TI3D), Ren is engineering optimal computational algorithms for drug discovery using the Ranger supercomputer and a large pool of known protein-ligand matches.

"We're testing and developing computational approaches that can best reproduce the experimental data of protein-ligand binding that has been reported in the literature," explained Ren.

The effort, one of the most comprehensive studies of protein-ligand interaction using all-atom molecular simulations, investigates a range of physical models: from implicit to explicit representation of water solvent, simple or elaborated electrostatic models, and various treatment of molecular flexibility.

Ren will evaluate the accuracy and efficiency of these methods by applying them to more than 200 complexes across 10 different protein families, and simulating their dynamics over long periods of time to provide a sufficient sampling of the molecular interaction. Ren believes the large trial size and longer molecular simulations will allow the researchers to test the protein binding algorithms more rigorously and determine the optimal methods for drug discovery.

Better Physics = Better Answers

Structure of benzamidine-bound trypsin and chemical structure of trypsin ligands studied (A-F).

The problem with the existing computational approaches used in the field is that they poorly represent physical reality, leading to simulations that do not reliably return correct compounds.

"In the old days, shortcuts were necessary to achieve speed. Researchers made approximations of physical models because the computations were too expensive," said Ren of the traditional docking methods. "We're basically adding that further layer of physics in order to get more accurate prediction."

These improved methods are implemented in the AMBER (Assisted Model Building with Energy Refinement) and TINKER molecular modeling packages. Ren and colleagues are experimenting with algorithms that use explicit or continuum methods to describe the solvent environment and account for entropic contribution to protein-ligand binding via molecular dynamics simulation.

A more sophisticated electrostatic model that treats the electronic polarization between ligand and environment, developed by Ren's group, is also under investigation.

"The promise of rapid, inexpensive computational drug discovery has thus far eluded scientists," said Michael Gonzales, Life Sciences Program Director at TACC. "Pengyu's work is an excellent example of how current advances in computing power are enabling scientists to take a fundamentally different approach to virtual drug discovery."

The greater physical reality of these algorithms is enabled by the massive computing power of TACC's parallel systems. "Having access to TACC, we're in a position to actually try out these methods," said Ren.

By testing these varied algorithms on 200 molecules, Ren ensures their effectiveness on a range of sample problems. The method that has the closest correlation to the experimentally determined results is deemed the most effective. And if the most successful method can be proven to work consistently, Ren believes chemists will adopt it, even if the computing cost is higher.

Pengyu Ren, assistant professor of biomedical engineering at The University of Texas at Austin

"Researchers like to use the minimum necessary computing effort when solving a problem," Ren said. "Unless you can show more accuracy and prove that it's worth the expense, they won't switch. That's what we're hoping to demonstrate."

Building Knowledge, Making Predictions

It's not all about methods and protocols for Ren and his collaborators. In turns out that basic aspects of protein-ligand binding are poorly understood, and it's not easy for laboratory experiments to show what's happening in these systems at the atomic level. Consequently, scientists must simulate the binding mechanisms in ultra-high resolution to account for the dynamics of every molecule in a large system.

"That's the scientific part," said Ren. "How do the ligands recognize the protein targets? What is the role of electrostatic interactions? What are the entropy and solvent effect on the binding? All that gives us the understanding that we can use to guide the design of ligands."

Along these lines, Ren is working with Dr. Stephen Martin (UT Chemistry) and Dr. Ron Elber (UT Biochemistry) to understand the fundamental relationship between ligand hydrophobicity, rigidity, and protein-ligand binding affinity from molecular simulations.

Ren is also involved in a joint project with Dr. Kevin Dalby (UT Medicinal Chemistry), where computation will guide the search for selective inhibitors for protein kinases, which are clinically relevant to cancer and other diseases.

Throughout the pharmaceutical industry and academia, researchers are looking for more effective methods that can predict promising compounds for a given protein target with minimal effort and maximum precision.

"If this works, it will improve our ability to design drug candidates that are more potent with fewer side-effects," said Ren. "But to make robust, accurate predictions, it's time to invest in the next generation of computational technologies for drug discovery."

Story Highlights

Due to the time and costs involved, computing is needed to speed drug discovery efforts. By creating virtual models of proteins and ligands, and automating the interaction process, chemists can screen possible compounds faster and at a lower cost.

Simple docking programs miss many of the most effective drug compounds; yet more accurate methods have been too expensive or complicated to employ.

Researchers from The University of Texas at Austin are using the Ranger supercomputer to benchmark improved drug discovery methods and to predict effective candidates for proteins involved in cancer and heart disease.


Faith Singer-Villalobos

Communications Manager | 512-232-5771

Aaron Dubrow

Science And Technology Writer

Jorge Salazar

Technical Writer/Editor | 512-475-9411