Drug-like density: a method of quantifying the "bindability" of a protein target based on a very large set of pockets and drug-like ligands from the Protein Data Bank.

One approach to estimating the "chemical tractability" of a candidate protein target where we know the atomic resolution structure is to examine the physical properties of potential binding sites. A number of other workers have addressed this issue. We characterize ~290,000 "pockets" from ~42,000 protein crystal structures in terms of a three parameter "pocket space": volume, buriedness, and hydrophobicity. A metric DLID (drug-like density) measures how likely a pocket is to bind a drug-like molecule. This is calculated from the count of other pockets in its local neighborhood in pocket space that contain drug-like cocrystallized ligands and the count of total pockets in the neighborhood. Surprisingly, despite being defined locally, a global trend in DLID can be predicted by a simple linear regression on log(volume), buriedness, and hydrophobicity. Two levels of simplification are necessary to relate the DLID of individual pockets to "targets": taking the best DLID per Protein Data Bank (PDB) entry (because any given crystal structure can have many pockets), and taking the median DLID over all PDB entries for the same target (because different crystal structures of the same protein can vary because of artifacts and real conformational changes). We can show that median DLIDs for targets that are detectably homologous in sequence are reasonably similar and that median DLIDs correlate with the "druggability" estimate of Cheng et al. (Nature Biotechnology 2007, 25, 71-75).

QSAR models for predicting the similarity in binding profiles for pairs of protein kinases and the variation of models between experimental data sets.

We propose a direct QSAR methodology to predict how similar the inhibitor-binding profiles of two protein kinases are likely to be, based on the properties of the residues surrounding the ATP-binding site. We produce a random forest model for each of five data sets (one in-house, four from the literature) where multiple compounds are tested on many kinases. Each model is self-consistent by cross-validation, and all models point to only a few residues in the active site controlling the binding profiles. While all models include the "gatekeeper" as one of the important residues, consistent with previous literature, some models suggest other residues as being more important. We apply each model to predict the similarity in binding profile to all pairs in a set of 411 kinases from the human genome and get very different predictions from each model. This turns out not to be an issue with model-building but with the fact that the experimental data sets disagree about which kinases are similar to which others. It is possible to build a model combining all the data from the five data sets that is reasonably self-consistent but not surprisingly, given the disagreement between data sets, less self-consistent than the individual models.

Spiroimidazolidinone NPC1L1 inhibitors. 1: Discovery by 3D-similarity-based virtual screening.

A series of spiroimidazolidinone NPC1L1 inhibitors was discovered by virtual screening of the Merck corporate sample repository using 3D-similarity-based screening. Selection of 330 compounds for testing in an in vitro NPC1L1 binding assay yielded six hits in six distinct chemical series. Follow-up 2D similarity searching yielded several sub- to low-micromolar leads; among these was spiroimidazolidinone 10, with an IC(50) of 2.5 microM. Compound 10 provided a useful scaffold to initiate a medicinal chemistry campaign.

Comparison of topological, shape, and docking methods in virtual screening.

Virtual screening benchmarking studies were carried out on 11 targets to evaluate the performance of three commonly used approaches: 2D ligand similarity (Daylight, TOPOSIM), 3D ligand similarity (SQW, ROCS), and protein structure-based docking (FLOG, FRED, Glide). Active and decoy compound sets were assembled from both the MDDR and the Merck compound databases. Averaged over multiple targets, ligand-based methods outperformed docking algorithms. This was true for 3D ligand-based methods only when chemical typing was included. Using mean enrichment factor as a performance metric, Glide appears to be the best docking method among the three with FRED a close second. Results for all virtual screening methods are database dependent and can vary greatly for particular targets.

Scoring of KDR kinase inhibitors: using interaction energy as a guide for ranking.

Within a congeneric series of ATP-competitive KDR kinase inhibitors, we determined that the IC(50) values, which span four orders of magnitude, correlated best with the calculated ligand-protein interaction energy using the Merck Molecular Force Field (MMFFs(94)). Using the ligand-protein interaction energy as a guide, we outline a workflow to rank order virtual KDR kinase inhibitors prior to synthesis. When structural information of the target is available, the ability to score molecules a priori can be used to rationally select reagents. Our implementation allows one to select thousands of readily available reagents, enumerate compounds in multiple poses and score molecules in the active site of a protein within a few hours. In our experience, virtual library enumeration is best used when a correlation between computed descriptors/properties and IC(50) or K (i) values has been established.

Dynamic control of allosteric antagonism of leukocyte function antigen-1 and intercellular adhesion molecule-1 interaction.

Leukocyte function associated antigen-1 (LFA-1) plays a critical role in T cell migration and has been recognized as a therapeutic target for immune disorders. Several classes of small molecule antagonists have been developed to block LFA-1 interaction with intercellular adhesion molecule-1 (ICAM-1). Recent structural studies show that the antagonists bind to an allosteric site in the I-domain of LFA-1. However, it is not yet clear how these small molecules work as antagonists since no significant conformational change is observed in the I-domain-antagonist complex structures. Here we present a computational study suggesting how these allosteric antagonists affect the dynamics of the I-domain. The lowest frequency vibrational mode calculated from an LFA-1 I-domain structure shows large scale "coil-down" motion of the C-terminal alpha7 helix, which may lead to the open form of the I-domain. The presence of an allosteric antagonist greatly reduces this motion of the alpha7 helix as well as other parts of the I-domain. Thus, our study suggests that allosteric antagonists work by eliminating breathing motion that leads to the open conformation of the I-domain.

Enhanced virtual screening by combined use of two docking methods: getting the most on a limited budget.

Flexible ligand docking is a routine part of a modern structure-based lead discovery process. As of today, there are quite a number of commercial docking programs that can be used to screen large databases (hundreds of thousands to millions of compounds). However, limiting factors such as the number of commercial software licenses needed to perform docking simultaneously on multiple processors ("software cost") and the relatively long time required per molecule to get good results ("quality-to-speed") should be taken into account when planning a large docking run. How can we optimize the efficiency of selecting lead candidates by docking, in respect to the quality of the results, search speed, and software cost? We present a combination of two methods, our "fast-free-approximate" in-house docking program and the "slow-costly-accurate" ICM-Dock, as an example of one solution to the problem. Our proposed protocol is illustrated by a series of virtual screening experiments aimed at identifying active compounds in the MDL Drug Data Report database. In more than half of the 20 cases examined, at least several actives per protein target were identified in approximately 24 hours per target.

Structural and functional modeling of human lysozyme reveals a unique nonapeptide, HL9, with anti-HIV activity.

We previously reported that lysozyme accounts for anti-HIV activity associated with the beta-core fraction of human chorionic gonadotropin [Lee-Huang, S., Huang, P. L., Sun, Y., Kung, H. F., Blithe, D. L. & Chen, H. C. (1999) Proc Natl Acad Sci U S A 96, 2678-81]. To define the structural and sequence requirements for anti-HIV activity, we carried out peptide fragmentation and activity mapping of human lysozyme. We identified two peptides that consist of 18 and 9 amino acids of human lysozyme (HL18 and HL9), corresponding to residues 98-115 and 107-115. HL18 and HL9 are potent inhibitors of HIV-1 infection and replication with EC(50)s of 50 to 55 nM, comparable to intact lysozyme. Scrambling the sequence or substitution of key arginine or tryptophan residues results in loss of antiviral activity. HL9, with the sequence RAWVAWRNR, is the smallest peptide we identified with full anti-HIV activity. It forms a pocket with its basic residues on the surface of the molecule. HL9 exists as an alpha-helix in native human lysozyme, in a region of the protein distinct from the muramidase catalytic site. Monte Carlo peptide folding energy minimizing simulation modeling and CD studies indicate that helical propensity does not correlate with antiviral activity. HL9 blocks HIV-1 viral entrance and replication, and modulates gene expression of HIV-infected cells, affecting pathways involved in survival, stress, TGFbeta, p53, NFkappaB, protein kinase C and hedgehog signaling.

Similarity to molecules in the training set is a good discriminator for prediction accuracy in QSAR.

How well can a QSAR model predict the activity of a molecule not in the training set used to create the model? A set of retrospective cross-validation experiments using 20 diverse in-house activity sets were done to find a good discriminator of prediction accuracy as measured by root-mean-square difference between observed and predicted activity. Among the measures we tested, two seem useful: the similarity of the molecule to be predicted to the nearest molecule in the training set and/or the number of neighbors in the training set, where neighbors are those more similar than a user-chosen cutoff. The molecules with the highest similarity and/or the most neighbors are the best-predicted. This trend holds true for narrow training sets and, to a lesser degree, for many diverse training sets and does not depend on which QSAR method or descriptor is used. One may define the similarity using a different descriptor than that used for the QSAR model. The similarity dependence for diverse training sets is somewhat unexpected. It appears to be greater for those data sets where the association of similar activities vs similar structures (as encoded in the Patterson plot) is stronger. We propose a way to estimate the reliability of the prediction of an arbitrary chemical structure on a given QSAR model, given the training set from which the model was derived.