YouTube Facebook LinkedIn Google+ Twitter Xinginstagram rss  

Focusing Drug Computation on Fragments

By Mike May

August 8, 2007 | Rather than screening complete molecules, many companies are turning to chemical fragments in their quest for new drugs. Fragments can be screened just like small molecules, running them through wet binding assays. In fact, some companies, such as ActiveSight and ChemBridge sell fragment libraries. But considering all the possible combinations and how they can be attached is almost undoable through conventional means.

In silico techniques offer growing potential to create and simulate various characteristics of virtual drug libraries, largely because the computational issues are more manageable with fragments. As Stephen Burley, CSO and senior VP research at SGX Pharmaceuticals, says: “In silico methods play a key role, especially with fragments, because there’s so much room to move in the modification of a fragment that the list of options is daunting.”

Jeff Wiseman, vice president technology and corporate development at Locus Pharmaceuticals, (See “Locus Focus,” Bio•IT World, Dec. 2002) notes four challenges for in silico drug discovery:

•           Force fields — the bonding energy between atoms — must be calculated. The current force field predictions, basically estimates of the actual forces, are good, but better ones are on the horizon.

•           A program must calculate a system’s interaction energy, or free energy, which consists of the enthalpy and entropy. “The force field gives enthalpy easily,” Wiseman says. Entropy requires a complicated simulation to sum interactions across the whole system, but it too can be calculated.

•           Current calculations assume a gas phase, but the real world is liquid. A good correction for that, Wiseman believes, is about one year away.

•           Finally, most calculations depend on a protein’s crystal structure. However, as Wiseman says, “But they move — a lot!”  Moreover, “As the size of the molecule increases, the degree of difficulty increases exponentially.” Programmers take on that problem with brute force, but it appears that today’s computing power is up to the task.

Despite these challenges, in silico techniques are pushing ahead in fragment-based drug discovery.

Force Field Forecasting
To simulate fragment-based drugs, the software company Schrödinger uses physics-based approaches that rely on quantum mechanics and accurate force fields. As Woody Sherman, director of applications science at Schrödinger, says, “That’s where the right answer lies — in quantum mechanics and making the right force field approximations.”

Nonetheless, treating large systems with quantum mechanics is impractical due to the computational complexity, and producing an accurate force field can be a challenge because it’s not always clear which parameters are inaccurate or missing from the existing force fields. To get a better handle on that, Schrödinger scientists are developing the next-generation force field. They screened a database of more than one million commercially available compounds to look for missing parameters in the current force-field simulations. As a next step, they are running high-level quantum mechanics to fill in the missing pieces.

Schrödinger scientists also apply quantum mechanics to protein-ligand interactions, such as fragment binding. Schrödinger scientists select a small region around the binding site and treat it quantum mechanically; the rest of the system gets simulated with a molecular-mechanics force field. “The idea,” says Sherman, “is that we treat the important part of the system as accurately as possible and rely on force fields derived from quantum mechanics to treat the rest of the system.” He adds, “Because fragments are small, we are able to treat more of the important part of the system with quantum mechanics.”

Other software companies rely heavily on experimental data. For example, Simulations Plus makes ClassPharmer, which can explore data from high-throughput screening to detect a specific fragment’s impact on activity. Then, says CEO Walt Woltosz, “Our ADMET Predictor can be used to examine a structure’s likely toxicity and many other properties.”

To test the power of this approach, Woltosz and his colleagues took a set of COX-2 inhibitors, broke them into substituent fragments, and simulated new combinations to create about 27,000 potential molecules. Because all molecules only use fragments that have been previously synthesized, synthetic feasibility is likely. ADMET Predictor was then used to provide an ADMET Risk score, which identified 70 or so low-risk compounds. That group included Celecoxib, an effective drug.

Despite the small size of fragments, they require extensive computation. David Lorber, product manager of pharmacophore modeling and database management at Accelrys, asks: “How does one deal with the very large number of molecules possible from combining fragments?” His answer involves Accelrys’ Discovery Studio and a good fragment library.

“Start with known drug molecules and generate pharmacophores,” says C. M. Venkatachalam, a fellow at Accelrys. “Then search a database to find fragments satisfying some of the pharmacophore features and join these, using the Accelrys joiner algorithm, to create new drug candidates.” This already limits the computations somewhat, because it considers only the fragments that satisfy certain features identified from the known drug. Venkatachalam says, “Starting with a known drug provides the advantage of restricting the space; but we are looking at other approaches to produce super-novel drug candidates.”

Putting Fragments to Work
To render the number of potential fragment modifications manageable, SGX scientists combine a variety of techniques. First, they screen a library of about 1,400 fragments — or scaffolds, as Burley prefers to call them — by using X-ray crystallography to see which ones bind to a target. That screening usually turns up 15 to 70 starting points that chemists cut to about five scaffolds. “We use in silico methods to explore chemical modifications of each scaffold,” Burley says.

Next, SGX chemists synthesize a subset of the modified fragments and run them through assays, such as IC50 measurements. Eventually, this process leads to 50-100 modified fragments for further testing. “The interplay between crystallography, enzyme assays, and in silico methods,” says Burley, “gives a ‘truer’ picture of the structure-activity relationship. You see how the small molecule binds to the target, the consequences of each synthetic change on the binding affinity, and that helps you decide what to do next in the lead-optimization process.”

SGX scientists used this technique against a Gleevec-resistant form of BCR-ABL, the target for treating chronic myeloid leukemia. The SGX team focused on the T315I mutation of BCR-ABL, which accounts for about 20 percent of the mutations that cause Gleevec resistance. Using its fragment-based approach, SGX discovered highly selective and specific inhibitors of wild-type and Gleevec-resistant BCR-ABL.

Few groups rely on in silico approaches alone. Instead, pharmaceutical scientists and software developers use various sources of information. But as Wiseman says, “The basic computational methods are solvable now. The issues for the future are more: How do you learn to use the methods effectively?” He adds, “We need to bring in people skilled in thinking about data and how to interpret it.”

Subscribe to Bio-IT World  magazine.

Click here to login and leave a comment.  


Add Comment

Text Only 2000 character limit

Page 1 of 1

For reprints and/or copyright permission, please contact Angela Parsons, 781.972.5467.