By throwing a wrench into biological machinery, medications (drugs) can alleviate many diseases, conditions, and ailments in the human body. Some drugs shut down enzymes that catalyze essential chemical reactions. Others interfere with communication between and within cells. Disrupting these critical functions can sabotage a pathogen or correct chemical imbalances in the body.
Knowing exactly which drugs or compounds will bind tightly to which receptors is key to drug discovery and development. If a molecule binds tightly and specifically to a target, it's much more likely to be a safe and effective drug - and avoid interacting with unintended targets and causing side effects. It's also critical that a drug binds tightly so that a small dose works well.
Molecular modeling and 'docking' methods are commonly used to predict the strength of association, or binding affinity, between a molecule and a receptor, by minimizing the energy of the overall system. Looking for the lowest overall energy configuration, however, is only one way to approach the problem. David Minh, an assistant professor of chemistry at the Illinois Institute of Technology, US, has recently derived a new and formally exact theory for calculating binding affinities.
Binding affinity tradeoffs: Time, accuracy, cost
In common docking simulations that look for the lowest system energy, Minh explains that the larger cell receptor is usually rigid, while the smaller molecular compound is moving around. This results in an approximate binding affinity. This type of docking, says Minh, "can be achieved with quick serial throughput, and is a routine way to screen for millions of compounds."
Out of millions of compounds, results may rank the top fifty. Of that fifty, two or three may actually have a high binding affinity - that is better than picking two or three compounds at random, but the technique is still not that accurate. Each failed binding experiment takes time, costs money, and pushes drug development further down the road.
In contrast, an alternative approach - free energy calculation - is more accurate, but computationally expensive. "These calculations try to account for movement in both the smaller molecule and the larger receptor," Minh says, and the results include a much more accurate binding affinity.
Because there are so many atoms moving about, free energy calculations are best suited to high-performance compute clusters with several thousand CPUs. Each CPU simulates a part of the molecular system and communicates frequently with the others.
The expense and resources required to generate free energy calculations make them much less common, albeit more accurate for predicting which molecules are likely to bind to which receptors. Schrödinger, a software and services provider for life sciences research, is backed financially by Bill Gates. According to Robert Abel, vice president of scientific development, the company has completed several thousand free energy calculations in total.
AlGDock: A hybrid approach
Minh's methodology is a hybrid between traditional docking and free energy calculation. "The theory shows that it is possible to derive the binding free energy from multiple simulations where the receptor is frozen," says Minh.
Each of the simulations where the receptor is frozen can be run on one CPU, in time frames ranging from one day to a week. That's perfect for running on Open Science Grid, adds Minh, because when you "simulate the binding free energy of the ligand to the frozen receptor, each calculation is completely independent; you can spread that out over many processes and get the binding free energy."
Minh's methodology should ultimately make it routine to run several thousand free energy calculations. "We've completed a couple thousand calculations already. It's still a method that's being developed, but statistically it's superior to docking." This approach has the potential to combine the speed of docking that uses rigid receptor structures with the accuracy of slower methods that allow full receptor flexibility.
Finding a molecule that will interfere with a chosen target isn't easy, to say the least. The number of possible small organic molecules is astronomical, and estimated to be around 1060. Solving these structures is like putting together a giant jigsaw puzzle and trying to create the missing piece. Even with the help of computers, the task is enormous. AlGDock could save experimentalists time, and ultimately mean finding potential drugs with fewer resources.