Discussion

Slide is an efficient database screening tool, which searches data sets of structures of more than 175 000 organic compounds within minutes, when using a small template, as we did in the case of the progesterone receptor screen, or within several hours, as shown for uracil-DNA glycosylase and dihydrofolate reductase, where we used a more general binding-site template. It accomplishes this by using an efficient multi-level hashing scheme to directly access triplets of feasible interaction points in the binding-site template, onto which triplets of ligand interaction centers are mapped. On one hand this is a straightforward way to compute a transformation of the ligand into the binding site, so that the ligand already makes three favorable interactions, and on the other hand it is also an efficient way to rule out infeasible compounds: all compounds that lack a set of three favorable interactions are discarded before attempting docking into the binding site, For the progesterone receptor with the very specific 6-point template, more than 163 000 compounds, i.e., 93% of the screening databases, never needed to be docked into the binding site

Figure 7. Another known inhibitor for DHFR found by Slide in the CSD: 2,4-dia-mino-5-( 1-adamantyl)-6-methylpyrirnidine (CSD entry FIRNID), which binds with higher affinity to DHFR than methotrexate [65]. Again, the pyrimidine ring is located in the targeted area of the binding site and makes one water-mediated interaction. The other water molecule located in that area (shown in Figure 6) was displaced by a polar amino group, resulting in no desolvation penalty. Note the hydrophobic complementarity of the side chains in contact with the adamantan (yellow indicates their initial conformations).

Figure 7. Another known inhibitor for DHFR found by Slide in the CSD: 2,4-dia-mino-5-( 1-adamantyl)-6-methylpyrirnidine (CSD entry FIRNID), which binds with higher affinity to DHFR than methotrexate [65]. Again, the pyrimidine ring is located in the targeted area of the binding site and makes one water-mediated interaction. The other water molecule located in that area (shown in Figure 6) was displaced by a polar amino group, resulting in no desolvation penalty. Note the hydrophobic complementarity of the side chains in contact with the adamantan (yellow indicates their initial conformations).

for this reason. For more general templates, like the 155-point template for uracil-DNA glycosylase, docking and conformational search were performed for more than 70 000 compounds (40% of the database).

Early in the development of SLIDE, we tried to reduce the complexity of the conformational search for the protein by using a rotamer library for the side chains, which had been done in docking approaches [31,32]. However, in the majority of cases all rotamers cause new collisions, and in a recent study it was shown that side chains close to ligand-binding sites tend to adopt non-rotameric conformations [58]. In most cases, including the examples described above, only minor rotations in both ligand and protein are necessary to generate a shape-complementary interface. These rotations are computed exactly by SLIDE, avoiding costly sampling of rotational angles.

The conformational search is the most computationally complex step of screening with Slide. Our model of flexibility is more realistic than that in docking or screening tools that only consider ligand flexibility, since ligand and protein flexibility are treated equally, and the mean-field optimization selects those rotations for resolving collisions that cause the minimal overall distortion for the complex. Full conformational search is not done for the ligand, but rather its database conformation is used as a starting conformation for docking. Since the structures for potential ligands are taken from crystal structures (CSD) or rule-derived models (NCI), they begin in a low-energy conformation. To deal with cases where the binding conformation of a ligand is very different from the database conformation, the database can be enriched by a series of low-energy conformers for screening [28,59].

Although our scoring function was empirically tuned based on published affinities for PDB complexes, we do not try to predict precise binding affinities in SLIDE. Several empirically derived scoring functions can be found in the literature [51,60-65]. Scoring functions sensitive to small conformation changes may not be appropriate for a screening tool like SLIDE, which cannot perform a conformational search for 100 000 or more ligand candidates. A sensitive scoring function is more appropriate in a fine-docking tool, which must predict differences in binding affinities for very similar conformations of a complex during the search. The scoring function in SLIDE is designed instead to rank the set of all potential ligands based on their complementarity. All examples described above were ranked within the top potential ligands for each target protein. SLIDE includes a web-based interface that enables the user to easily browse through the potential ligands and visualize Slide's docking.

The inclusion of binding-site solvation is in accordance with our models of induced fit and scoring. The positions of water molecules in the binding site from the crystal structure of the target protein are analyzed, and those predicted as conserved by Consolv are kept. In contrast to a method that precomputes several favorable water positions prior to docking, then picks the best positions to fill gaps between the molecules [43], Slide starts with 'real' water molecules and shifts them when they collide with ligand atoms. As in the conformational search, the idea is to start with a reasonable configuration and make only minimal changes, as necessary. If the collision of a water molecule cannot be resolved, the water is displaced and a desolvation penalty term is only applied when a lost hydrogen bond is not replaced by a corresponding protein-ligand interaction.

While Slide's docking procedure must be very quick, rather than comprehensive, in order to screen a large number of molecules, its inclusion of protein flexibility and solvation gives SLIDE advantages over other docking procedures. Because of the fast screening time, Slide can be used to search very large compound databases for the discovery of novel lead structures, and due to distinguishing a rigid anchor fragment for each screened molecule attached to flexible side chains, it will be straightforward to extend SLIDE for combinatorial screening.

0 0

Post a comment