Major Limitation of Most Existing Methods. Treating the receptor as fixed during docking is generally a necessary evil. In many docking programs, the receptor is fixed for all calculations. The need to account for the dynamic behavior of a receptor has long been recognized as a complicating factor in computational drug design. The energy landscape of most proteins and nucleic acid are frequently described in terms of a folding funnel in which there are many highly unfavorable states that collapse via multiple routes entailing favorable folded states generally arriving at a single native structure, which still may possess some conformational variability at ambient temperature. A single structure, even the weighted average provided by a crystal structure or NMR structure, may not adequately describe these substates required for ligand recognition. It is important to note that a conformational funnel and the states that it represents are condition-dependent. Altering the conditions (ionic strength, pH, temperature) changes details of the funnel. One of the most important points to consider is that introducing a ligand into the system also changes the environment. It may affect the most populated state of the receptor; such a case would correspond to an “induced-fit” of ligand and receptor. The macromolecule most likely exists in a full complement of conformations: most in the native state, some in the induced-fit state, and some in other states. If the ligand binds preferentially to the induced-fit state with sufficiently favorable free energy change, i.e., greater than the free energy difference between the native and induced-fit states of the macromolecule, the average structure of the macromolecule will change. If the ligand is capable of multiple binding modes to a macromolecule (binding to different folded forms or multiple binding sites of a single form), it may be necessary to account for these additional states to predict affinity properly.
Many docking algorithm build the drugs “in situ” in a pocket without checking how the drugs bind there kinetically. Studies in our lab using MORDOR show that ~20% of the drugs found with these techniques cannot reach the pocket. While the “fix” of building “in situ” is certainly an appropriate response to get calculated structures to match experimental, it illustrates the difficulty of adequately testing a potential ligand for binding when the receptor changes its structure in response to the ligand’s presence.
MORDOR docking software was primarily designed for accuracy rather than rapidity of docking. (We will subsequently streamline MORDOR, so it can run as fast as possible to achieve its goal.) Most importantly, MORDOR allows induced-fit. The ligand is able to displace bases, and it can create a new bulge or stabilizing/ordering loop of the RNA. For that purpose, MORDOR implements a novel algorithm for conformational space search that allows an efficient systematic search during docking. This approach combines molecular minimization or molecular dynamics simulation with a driving force that moves the ligand. Starting from a random position of the ligand around the receptor, the ligand explores the receptor surface by an additional root-mean-square-deviation type of force (Path Exploration With Distance Constraints or PEDC method (see publications), which constrains the drug to explore the conformational space following a low energy pathway. The ligand explores and identifies all potential binding pockets as shown above.
One of the most important aspects of this approach is that we naturally take into account "induced-fit" when the drug binds to the receptor. Movements of side-chains, regions, or even whole domains may be permitted by the docking procedure. In particular, the presumed binding pocket may be allowed to change its shape to a certain extent for a better binding interaction. This allows us to include in our scoring function the energy change of the macromolecule while changing its conformation due to the drug interaction. This is very important and has been neglected until now in drug design. Our preliminary data show that contributions of the target induced-fit energy change could make the difference for predicting the correct binding energy. The second major advantage of using MORDOR is that we have a complete surface map of where the ligands may bind. That enables assessment of drug specificity
Flexible Docking with MORDOR. Studying induced fit with MORDOR is particularly important in docking protein and particularly nucleic acids. In fact, nucleic acid-drug complex structures indicate that drugs do not generally bind a canonical form of RNA or DNA. On the contrary, known structures of drugs bound to nucleic acids indicate that very often, the drugs displace bases thus provoking a local reorganization of the nucleic acid. Flexible docking for RNA will greatly enhance the probability of success in a drug discovery program.
Docking Procedure. We have created and currently use a database of ~400,000 – 500,000 compounds incorporating the Available Chemicals Directory (ACD) (≥200,000 compounds), the National Cancer Institute directory (~200,000 compounds), and compounds from commercial sources made available to us (Maybridge, Comgenex, TRC). AMBER/AnteChamber atom types were assigned based on local connectivity. The partial atom charge was calculated using the AM1-BCC correction factor.
We first perform
rigid docking using DOCK
effectively to reduce the size of libraries, while retaining the
potentially good drug candidates. DOCK is fast and convenient
to select 10,000 compounds from the database. It uses spheres that are supposed to represent the best
pocket on the receptor to fit the ligand. The spheres are usually
generated using accessible surface criteria, where a buried pocket
inside the macromolecule seems to be a place of choice. However, this
rapid approach uses a rigid structure, which precludes discovery of a
pocket made by the receptor as it alters its conformation due to the
presence of the ligand. To find new hot spots, we developed a procedure
using MORDOR to calculate the binding interaction energy map between a
receptor and a dummy sphere (given a VDW radius and a charge). The
dummy sphere is placed and moved along a grid that covers the entire
receptor. Both receptor and dummy sphere are flexible and are minimized
simultaneously. The result is an interaction energy map between the
receptor and the dummy sphere that identifies and finds interesting new
hotspots to interact with ligands. This procedure was able to identify
all known pockets in our training test set among the top ten spheres,
i.e., those that exhibited the best interaction energy with the
receptor. Sphere generation using MORDOR is illustrated above.
For present purposes of development, at the last stage of docking, the top 1000 compounds will be selected and investigated further for additional criteria: entropy term in the scoring function, drug specificity estimation by checking if the drugs can also bind somewhere else to the RNA surface, pathway of the ligand to reach the pocket from the solvent to assess whether it is kinetically favorable.Features of MORDOR.
Development. We are working on several aspects of docking.
MORDOR needs a very fast computational resource. It quickly becomes a
very slow process if screening a million compounds from vendor
databases or in-house libraries. It takes about 1-10 CPU-hr per complex
to generate the PEDC-driven pathway of binding using our current small
cluster. The proposed new cluster makes it possible to screen more than
10,000 compounds with MORDOR in its current non-optimized form.
We are also developing a new, particularly fast procedure using MORDOR
to pre-screen the database (instead of using DOCK), by predicting
binding affinity rapidly and with a reasonable level of accuracy.
We need to make improvements in the accuracy of the scoring function to enhance the reliability of discriminating correctly docked from misdocked conformations by adding: