Drug Development

Flexscreen performs fully automated in-silico screening of a large 3D database of ligands against a structurally resolved protein receptor. Each ligand of the database is docked against the receptor with the stochastic tunneling method using an all-atom representation of both ligand and receptor. Both ligand and receptor can change their conformation in the docking process. Using an all-atom scoring function based on physical principles, the affinity of the each ligand to the receptor is evaluated. The ligands with the best affinity are selected as lead candidates for drug development. Using our secure, high-performace computational platform, we presently screen the open ligand database of the National Cancer Institute (USA) which contains about 250,000 ligands, but other databases can also be used.

The specificity and selectivity of FlexScreen in lead selection is rooted in:

Lead Screening: Many modern strategies for drug design focus on the design of artificial ligands for a specific protein receptor. Compounds that have a high affinity to a protein involved in a disease may function as drugs, provided they are nontoxic,  bioactive and specific. Once an appropriate protein target has been chosen, the first important step is the identification of compounds (leads) that can bind to the protein with sufficient affinity. These leads are subjected to a refinement and selection process that can ultimately lead from in-vitro, to animal and finally to human clinical studies.  On the basis of known protein structures and established principles protein-ligand association, computational methods are increasingly used to identify suitable ligands for a given receptor. In lead-screeing all compounds in a large database are ranked on the basis of the computed affinity. This approach, known as in-silico high throughput screening (IS-HTS), selects the molecules with the highest predicted affinities as candidates for further experimental validation and development.

Present IS-HTS methods must compromise between accuracy and computational cost. Scoring functions are used to assign an affinity to each possible geometric arrangement of the receptor-ligand complex.  Optimization methods are employed to determine the best possible relative arrangement (binding mode) of ligand and receptor. The value of the scoring function for this complex is then used as an estimate of the affinity. The accuracy scoring function and the variability and the efficiency of the search method crucially influences the quality of the affinity calculation. It is known that many ligands attain their specificity because of their geometric complementarity (key-lock-principle). Since both receptor and ligand change their conformation in the association process, this flexibility must be accounted for in the search process to determine the ideal binding mode. While ligand flexibility is now routinely treated in most modern IS-HTS methods,  FlexScreen is the first approach that can account for continuous flexibility of the most important parts of  the protein that stabilize the receptor-ligand complex. This advance is rooted in the use of an efficient novel optimization method (stochastic tunneling approach), that permits an efficient solution of the complicated search for the optimal position. Because many important interactions that stabilize the receptor-ligand complex are physically short ranged, the accurate atomistic treatment of the entire complex permits the use of more detailed scoring functions to evaluate the affinity of the resulting receptor-ligand complex.

There are two important tests of the accuracy and reliability of an IS-HTS approach:

Reproduction of experimental binding modes: Since experimental data for many receptor ligand complexes is available, the accurate reproduction of such known complexes is an important necessary condition for a successful IS-HTS strategy. We have recently compared the accuracy of the binding modes of four different IS-HTS methods for the high-resolution subset of the Astex/CDC reference data set.  FlexScreen predicted the experimental binding modes with a median error of 0.76 Å, compared to errors of more than 1.2 Å for the other methods.  For 65 of 82 complexes more than 80% of the screening runs determined the binding mode to better than the exp­erimental accuracy (see figure: experiment (think) vs. 10 predicted modes (thin)). Comparing the data on a case-by-case basis FlexScreen finds a better binding mode in 65%, 80% and 85% of the cases in compa­rison with Glide, Gold and FlexX respectively. FlexScreen thus performs best by a significant margin in comparison to the other methods; its median error is comparable to the experimental resolution (2 Å).

Database Enrichment: The second test is used to asses the degree of database enrichment, i.e. the ability of an IS-HTS approach to select known high-affinity ligands in a large database of compe­ting molecules. We have investigated FlexScreen's ability to select ten known ligands to tymidine kinase from a randomly selected subset of ten thousand ligands of the NCI database. When we perform the test disabling FlexScreen’s ability to treat receptor flexibility, we obtain the result shown on the left. The horizontal axis shows the computed affinity for each ligand while the vertical axis counts the number of ligands found for this affinity. The known ligands are shown in yellow, the width of the curves illustrates the error in the affinity estimate. The left figure demonstrates that some ligands are assigned a very high relative affinity, such ligands would be selected for experimental study. Other ligands, placed in the center or at the bottom of the distribution would be clearly missed, three ligands did not dock at all.  On the right hand figure we have enabled receptor flexibility, which increases the cost by roughly a factor of five. Now all ligands dock and the enrichment rate, which measures the quality of the entire investigation,  increases by almost 50% in comparison with the rigid-receptor scan. In addition, 40% of the top 500 ligands in the flexible receptor screen, we not selected in the rigid receptor screen. These results demonstrate the superiority of IS-HTS screening tools, such as FlexScreen,  that can exploit continuous receptor flexibility in routine applications.

Histogram of a screen against the thymidin kinase receptor, the experimentally known substrates are shown in yellow.

What is unique in FlexScreen: Faster docking methods permit the use of more detailed models for ligands and receptor. This improves the accuracy of the affinity calculation and thus the quality of the lead candidates. We recently developed the stochastic tunneling method as a novel and very efficient generic optimization technique, which proved superior to competing techniques in a number of applications, including protein folding and receptor ligand docking. The superior speed of this method permits the use of high resolution all-atom models for lead screening with present day computational resources. In the last decade in-silico screening has moved from 2D pharmacophore models to 3D docking using first rigid, now flexible ligands in the screening process. With the additional incorporation of receptor flexibility at the all-atom level, FlexScreen offers a unique docking approach at the cutting edge of this technology. We use our advantage in accuracy to rationally design superior all-atom scoring functions, where we optimize the representation of the physical interactions between receptor and ligand.

For further information please contact Dr. W. Wenzel, Research Center Karlsruhe, Institute for Nanotechnology, P.O. Box 3640, D-76021 Karlsruhe, Germany, +49-72747-82-6386, wolfgang wenzelMph0∂kit edu,