A novel strategy for improving ligand selectivity in receptor-based drug design

Peptide science: A "rule model" for new generations of peptidomimetics

Peptides have been heavily investigated for their biocompatible and bioactive properties. Though a wide array of functionalities can be introduced by varying the amino acid sequence or by structural constraints, properties such as proteolytic stability, catalytic activity, and phase behavior in solution are difficult or impossible to impart upon naturally occurring α-L-peptides. To this end, sequence-controlled peptidomimetics exhibit new folds, morphologies, and chemical modifications that create new structures and functions. The study of these new classes of polymers, especially α-peptoids, has been highly influenced by the analysis, computational, and design techniques developed for peptides. This review examines techniques to determine primary, secondary, and tertiary structure of peptides, and how they have been adapted to investigate peptoid structure. Computational models developed for peptides have been modified to predict the morphologies of peptoids and have increased in accuracy in recent years. The combination of in vitro and in silico techniques have led to secondary and tertiary structure design principles that mirror those for peptides. We then examine several important developments in peptoid applications inspired by peptides such as pharmaceuticals, catalysis, and protein-binding. A brief survey of alternative backbone structures and research investigating these peptidomimetics shows how the advancement of peptide and peptoid science has influenced the growth of numerous fields of study. As peptide, peptoid, and other peptidomimetic studies continue to advance, we will expect to see higher throughput structural analyses, greater computational accuracy and functionality, and wider application space that can improve human health, solve environmental challenges, and meet industrial needs. STATEMENT OF SIGNIFICANCE: Many historical, chemical, and functional relations draw a thread connecting peptides to their recent cognates, the "peptidomimetics". This review presents a comprehensive survey of this field by highlighting the width and relevance of these familial connections. In the first section, we examine the experimental and computational techniques originally developed for peptides and their morphing into a broader analytical and predictive toolbox. The second section presents an excursus of the structures and properties of prominent peptidomimetics, and how the expansion of the chemical and structural diversity has returned new exciting properties. The third section presents an overview of technological applications and new families of peptidomimetics. As the field grows, new compounds emerge with clear potential in medicine and advanced manufacturing.

GRID and docking analyses reveal a molecular basis for flavonoid inhibition of Src family kinase activity

Flavonoids reduce cardiovascular disease risk through anti-inflammatory, anti-coagulant and anti-platelet actions. One key flavonoid inhibitory mechanism is blocking kinase activity that drives these processes. Flavonoids attenuate activities of kinases including phosphoinositide-3-kinase, Fyn, Lyn, Src, Syk, PKC, PIM1/2, ERK, JNK and PKA. X-ray crystallographic analyses of kinase-flavonoid complexes show that flavonoid ring systems and their hydroxyl substitutions are important structural features for their binding to kinases. A clearer understanding of structural interactions of flavonoids with kinases is necessary to allow construction of more potent and selective counterparts. We examined flavonoid (quercetin, apigenin and catechin) interactions with Src family kinases (Lyn, Fyn and Hck) applying the Sybyl docking algorithm and GRID. A homology model (Lyn) was used in our analyses to demonstrate that high-quality predicted kinase structures are suitable for flavonoid computational studies. Our docking results revealed potential hydrogen bond contacts between flavonoid hydroxyls and kinase catalytic site residues. Identification of plausible contacts indicated that quercetin formed the most energetically stable interactions, apigenin lacked hydroxyl groups necessary for important contacts and the non-planar structure of catechin could not support predicted hydrogen bonding patterns. GRID analysis using a hydroxyl functional group supported docking results. Based on these findings, we predicted that quercetin would inhibit activities of Src family kinases with greater potency than apigenin and catechin. We validated this prediction using in vitro kinase assays. We conclude that our study can be used as a basis to construct virtual flavonoid interaction libraries to guide drug discovery using these compounds as molecular templates.

Identification of binding specificity-determining features in protein families

We present a new approach for identifying features of ligand-protein binding interfaces that predict binding selectivity and demonstrate its effectiveness for predicting kinase inhibitor specificity. We analyzed a large set of human kinases and kinase inhibitors using clustering of experimentally determined inhibition constants (to define specificity classes of kinases and inhibitors) and virtual ligand docking (to extract structural and chemical features of the ligand-protein binding interfaces). We then used statistical methods to identify features characteristic of each class. Machine learning was employed to determine which combinations of characteristic features were predictive of class membership and to predict binding specificities and affinities of new compounds. Experiments showed predictions were 70% accurate. These results show that our method can automatically pinpoint on the three-dimensional binding interfaces pharmacophore-like features that act as "selectivity filters". The method is not restricted to kinases, requires no prior hypotheses about specific interactions, and can be applied to any protein families for which sets of structures and ligand binding data are available.

Deciphering the Arginine-binding preferences at the substrate-binding groove of Ser/Thr kinases by computational surface mapping

Protein kinases are key signaling enzymes that catalyze the transfer of γ-phosphate from an ATP molecule to a phospho-accepting residue in the substrate. Unraveling the molecular features that govern the preference of kinases for particular residues flanking the phosphoacceptor is important for understanding kinase specificities toward their substrates and for designing substrate-like peptidic inhibitors. We applied ANCHORSmap, a new fragment-based computational approach for mapping amino acid side chains on protein surfaces, to predict and characterize the preference of kinases toward Arginine binding. We focus on positions P−2 and P−5, commonly occupied by Arginine (Arg) in substrates of basophilic Ser/Thr kinases. The method accurately identified all the P−2/P−5 Arg binding sites previously determined by X-ray crystallography and produced Arg preferences that corresponded to those experimentally found by peptide arrays. The predicted Arg-binding positions and their associated pockets were analyzed in terms of shape, physicochemical properties, amino acid composition, and in-silico mutagenesis, providing structural rationalization for previously unexplained trends in kinase preferences toward Arg moieties. This methodology sheds light on several kinases that were described in the literature as having non-trivial preferences for Arg, and provides some surprising departures from the prevailing views regarding residues that determine kinase specificity toward Arg. In particular, we found that the preference for a P−5 Arg is not necessarily governed by the 170/230 acidic pair, as was previously assumed, but by several different pairs of acidic residues, selected from positions 133, 169, and 230 (PKA numbering). The acidic residue at position 230 serves as a pivotal element in recognizing Arg from both the P−2 and P−5 positions.

Protein kinases are key signaling enzymes and major drug targets that catalyze the transfer of phosphate group to a phospho-accepting residue in the substrate. Unraveling molecular features that govern the preference of kinases for particular residues flanking the phosphoacceptor (substrate consensus sequence, SCS) is important for understanding kinase-substrates specificities and for designing peptidic inhibitors. Current methods used to predict this set of essential residues usually rely on linking between experimentally determined SCSs to kinase sequences. As such, these methods are less sensitive when specificity is dictated by subtle or kinase-unique sequence/structural features. In this study, we took a different approach for studying kinases specificities, by applying a new fragment-based method for mapping amino acid side chains on protein surfaces. We predicted and characterized the preference of Ser/Thr kinases toward Arginine binding, using the unbound kinase structures. The method produced high quality predictions and was able to provide novel insights and interesting departures from the prevailing views regarding the specificity-determining elements governing specificity toward Arginine. This work paves the way for studying the kinase binding preferences for other amino acids, for predicting protein-peptide structures, for facilitating the design of novel inhibitors, and for re-engineering of kinase specificities.

Neglected disease - african sleeping sickness: recent synthetic and modeling advances

Human African Trypanosomiasis (HAT) also called sleeping sickness is caused by subspecies of the parasitic hemoflagellate Trypanosoma brucei that mostly occurs in sub-Saharan Africa. The current chemotherapy of the human trypanosomiases relies on only six drugs, five of which have been developed more than 30 years ago, have undesirable toxic side effects and most of them show drug-resistance. Though development of new anti-trypanosomal drugs seems to be a priority area research in this area has lagged far behind. The given review mainly focus upon the recent synthetic and computer based approaches made by various research groups for the development of newer anti-trypanosomal analogues which may have improved efficacy and oral bioavailability than the present ones. The given paper also attempts to investigate the relationship between the various physiochemical parameters and anti-trypanosomal activity that may be helpful in development of potent anti-trypanosomal agents against sleeping sickness.

Ligand binding site superposition and comparison based on Atomic Property Fields: identification of distant homologues, convergent evolution and PDB-wide clustering of binding sites

A new binding site comparison algorithm using optimal superposition of the continuous pharmacophoric property distributions is reported. The method demonstrates high sensitivity in discovering both, distantly homologous and convergent binding sites. Good quality of superposition is also observed on multiple examples. Using the new approach, a measure of site similarity is derived and applied to clustering of ligand binding pockets in PDB.

QSAR Studies on andrographolide derivatives as α-glucosidase inhibitors

Andrographolide derivatives were shown to inhibit α-glucosidase. To investigate the relationship between activities and structures of andrographolide derivatives, a training set was chosen from 25 andrographolide derivatives by the principal component analysis (PCA) method, and a quantitative structure-activity relationship (QSAR) was established by 2D and 3D QSAR methods. The cross-validation r² (0.731) and standard error (0.225) illustrated that the 2D-QSAR model was able to identify the important molecular fragments and the cross-validation r² (0.794) and standard error (0.127) demonstrated that the 3D-QSAR model was capable of exploring the spatial distribution of important fragments. The obtained results suggested that proposed combination of 2D and 3D QSAR models could be useful in predicting the α-glucosidase inhibiting activity of andrographolide derivatives.

Structure-based approach for the study of estrogen receptor binding affinity and subtype selectivity

Estrogens exert important physiological effects through the modulation of two human estrogen receptor (hER) subtypes, alpha (hERalpha) and beta (hERbeta). Because the levels and relative proportion of hERalpha and hERbeta differ significantly in different target cells, selective hER ligands could target specific tissues or pathways regulated by one receptor subtype without affecting the other. To understand the structural and chemical basis by which small molecule modulators are able to discriminate between the two subtypes, we have applied three-dimensional target-based approaches employing a series of potent hER-ligands. Comparative molecular field analysis (CoMFA) studies were applied to a data set of 81 hER modulators, for which binding affinity values were collected for both hERalpha and hERbeta. Significant statistical coefficients were obtained (hERalpha, q(2) = 0.76; hERbeta, q(2) = 0.70), indicating the internal consistency of the models. The generated models were validated using external test sets, and the predicted values were in good agreement with the experimental results. Five hER crystal structures were used in GRID/PCA investigations to generate molecular interaction fields (MIF) maps. hERalpha and hERbeta were separated using one factor. The resulting 3D information was integrated with the aim of revealing the most relevant structural features involved in hER subtype selectivity. The final QSAR and GRID/PCA models and the information gathered from 3D contour maps should be useful for the design of novel hER modulators with improved selectivity.

Structure-based chemogenomics: analysis of protein family landscapes

Analysis of the three-dimensional structures of protein ligand complexes provides valuable insight into both the common interaction patterns within a target family and the discriminating features between the different members of a target family. Knowledge of the common interaction patterns helps to design target family focused chemical libraries for hit finding, while the discriminating features can be exploited to optimize the selectivity profile of a lead compound against particular member of a target family. Herein, we review the computational tools which have been developed to analyze crystal structures of members of a target family.

Synthetic non-peptide mimetics of alpha-helices

Proteins in nature fold into native conformations in which combinations of peripherally projected aliphatic, aromatic and ionic functionalities direct a wide range of properties. Alpha-helices, one of the most common protein secondary structures, serve as important recognition regions on protein surfaces for numerous protein-protein, protein-DNA and protein-RNA interactions. These interactions are characterized by conserved structural features within the alpha-helical domain. Rational design of structural mimetics of these domains with synthetic small molecules has proven an effective means to modulate such protein functions. In this tutorial review we discuss strategies that utilize synthetic small-molecule antagonists to selectively target essential protein-protein interactions involved in certain diseases. We also evaluate some of the protein-protein interactions that have been or are potential targets for alpha-helix mimetics.

COMPARISON OF METHODS FOR THE PREDICTION OF THE METABOLIC SITES FOR CYP3A4-MEDIATED METABOLIC REACTIONS

Predictions of the metabolic sites for new chemical entities, synthesized or only virtual, are important in the early phase of drug discovery to guide chemistry efforts in the synthesis of new compounds with reduced metabolic liability. This information can now be obtained from in silico predictions, and therefore, a thorough and unbiased evaluation of the computational techniques available is needed. Several computational methods to predict the metabolic hot spots are emerging. In this study, metabolite identification using MetaSite and a docking methodology, GLUE, were compared. Moreover, the published CYP3A4 crystal structure and computed CYP3A4 homology models were compared for their usefulness in predicting metabolic sites. A total of 227 known CYP3A4 substrates reported to have one or more metabolites adding up to 325 metabolic pathways were analyzed. Distance-based fingerprints and four-point pharmacophore derived from GRID molecular interaction fields were used to characterize the substrate and protein in MetaSite and the docking methodology, respectively. The CYP3A4 crystal structure and homology model with the reactivity factor enabled achieved a similar prediction success (78%) using the MetaSite method. The docking method had a relatively lower prediction success (approximately 57% for the homology model), although it still may provide useful insights for interactions between ligand and protein, especially for uncommon reactions. The MetaSite methodology is automated, rapid, and has relatively accurate predictions compared with the docking methodology used in this study.

Classification and comparison of ligand-binding sites derived from grid-mapped knowledge-based potentials

We describe the application of knowledge-based potentials implemented in the MOE program to compare the ligand-binding sites of several proteins. The binding probabilities for a polar and a hydrophobic probe are calculated on a grid to allow easy comparison of binding sites of superimposed related proteins. The method is fast and simple enough to simultaneously use structural information of multiple proteins of a target family. The method can be used to rapidly cluster proteins into subfamilies according to the similarity of hydrophobic and polar fields of their ligand-binding sites. Regions of the binding site which are common within a protein family can be identified and analysed for the design of family-targeted libraries or those which differ for improvement of ligand selectivity. The field-based hierarchical clustering is demonstrated for three protein families: the ligand-binding domains of nuclear receptors, the ATP-binding sites of protein kinases and the substrate binding sites of proteases. More detailed comparisons are presented for serine proteases of the chymotrypsin family, for the peroxisome proliferator-activated receptor subfamily of nuclear receptors and for progesterone and androgen receptor. The results are in good accordance with structure-based analysis and highlight important differences of the binding sites, which have been also described in the literature.

A chemometric study of phosphodiesterase 5 inhibitors

This work presents a chemometric classification for a set of phosphodiesterase 5 inhibitors, based on a pattern recognition method widely used in quantitative structure-activity (QSAR) studies, hierarchical cluster analysis (HCA) and principal component analysis (PCA), aiming to access the most relevant structural and physicochemical variables related to phosphodiesterase 5 inhibition and to quantify the similarity of the structures within the set of inhibitors. Our model is capable of classifying a test set of 26 known phosphodiesterase 5 inhibitors in terms of similarity, the results being consistent with published experimental data.

Matrix Metalloproteinase Target Family Landscape: A Chemometrical Approach to Ligand Selectivity Based on Protein Binding Site Analysis

To gain insight into the structural determinants for the matrix metalloproteinase (MMP) family, we characterized the binding sites of 56 MMP structures and one TACE (tumor necrosis factor alpha converting enzyme) structure using molecular interaction fields (MIFs). These MIFs were produced by two approaches: the GRID force field and the knowledge-based potential DrugScore. The subsequent statistical analysis using consensus principal component analysis (CPCA) for the entire binding site and each subpockets revealed both approaches to encode similar information about discriminating regions. However, the relative importance of the probes varied between both approaches. The CPCA models provided the following ranking of the six subpockets based on the opportunity for selective interactions with different MMPs: S1' > S2, S3, S3' > S1, S2'. The interpretation of these models agreed with experimental binding modes inferred from crystal structures or docking.

Separation of the strength and selectivity of the microbiological effect of synthetic dyes by spectral mapping technique

The growth inhibitory effect of 30 synthetic dyes on 22 bacteria (test organisms) belonging to various taxonomic groups was determined. The strength (potency) and selectivity of the biological effect were separated by the spectral mapping technique, reducing the dimensionality of the selectivity maps to two by the nonlinear mapping technique. The relationship between biological effect and physicochemical parameters of dyes was elucidated by stepwise regression analysis. It has been established that the strength of the effect of anthracene and trityl derivatives was higher than that of azobenzene dyes and significantly depended on the hydrophobicity of the compound. The selectivity of the effect also depended on hydrophobicity and on the nonpolar unsaturated surface area of the dyes. Gram negative and Gram positive bacteria differed in the strength and selectivity of their response to dyes indicating the marked impact of the taxonomical position on the response. Contrary to other multivariate mathematical statistical methods biological activity may be divided by SPM into potency and selectivity values, therefore, application of the technique in future QSAR studies is highly recommended.

Protein Alpha Shape Similarity Analysis (PASSA): a new method for mapping protein binding sites. Application in the design of a selective inhibitor of tyrosine kinase 2

We have developed a method that we have called Protein Alpha Shape Similarity Analysis (PASSA), that identifies interaction sites that can be utilised to achieve selectivity towards a protein. We have shown that this method is able to identify residues of tyrosine kinases that interact with known selective inhibitors using the following test cases: Abelson (Abl) kinase in complex with STI-571 and Janus kinase 2 (Jak2) in complex with AG-490. The 3D structures of the tyrosine kinase domains of Tyrosine kinase 2 (Tyk2) and Jak2 have been predicted by homology modelling. Computational docking of AG-490 and a set of tyrphostins known not to inhibit Jak2 indicated that our homology models are able to separate inhibitors from non-inhibitors. PASSA has also been used to identify unique properties of Tyk2. According to our results, interactions with hydrogen acceptors and donors on the following residues can be utilised to achieve selectivity towards Tyk2: Y955, E1053, D1062 and S1063. These residues are placed close to non-conserved hydrophobic pockets. The PASSA results, together with results from Multiple Copy Simultaneous Search (MCSS) were used to suggest functional groups of a selective Tyk2 inhibitor.

A simple method for visualizing the differences between related receptor sites

Pastor and Cruciani [J. Med. Chem. 38 (1995) 4637] and Kastenholz et al. [J. Med. Chem. 43 (2000) 3033] pioneered methods for comparing related receptors, with the ultimate goal of designing selective ligands. Such methods start with a reasonable superposition of high-resolution three-dimensional (3D) structures of the receptors. Next, molecular field maps are calculated for each receptor. Then the maps are analyzed to determine which map features are correlated with a particular subset of receptors. We present a method FLOGTV, based on the trend vector paradigm [J. Chem. Inf. Comput. Sci. 25 (1985) 64] to perform the analysis. This is mathematically simpler than the GRID/CPCA method of Kastenholz et al. and allows for the simultaneous comparison of many receptor structures. Also, the trend vector paradigm provides a method of selecting isopotential contours that are well above "noise". We demonstrate the method on four examples: HIV proteases versus two-domain acid proteases, thrombin versus trypsin and factor Xa, bacterial dihydrofolate reductases (DHFRs) versus vertebrate DHFRs, and P38 versus ERK protein kinases.

Structural classification of protein kinases using 3D molecular interaction field analysis of their ligand binding sites: target family landscapes

Protein kinases are critical components of signaling pathways and trigger various biological events. Several members of this superfamily are interesting targets for novel therapeutic approaches. All known eukaryotic protein kinases exhibit a conserved catalytic core domain with an adenosine 5'-triphosphate (ATP) binding site, which often is targeted in drug discovery programs. However, as ATP is common to kinases and other proteins, specific protein-ligand interactions are crucial prerequisites for valuable ATP site-directed ligands. In the present study, a set of 26 X-ray structures of eukaryotic protein kinases were classified into subfamilies with similar protein-ligand interactions in the ATP binding site using a chemometrical approach based on principal component analysis (PCA) and consensus PCA. This classification does not rely on protein sequence similarities, as descriptors are derived from three-dimensional (3D) binding site information only computed using GRID molecular interaction fields. The resulting classification, which we refer to as "target family landscape", lead to the identification of common binding pattern and specific interaction sites for particular kinase subfamilies. Moreover, those findings are in good agreement with experimental selectivity profiles for a series of 2,6,9-substituted purines as CDK inhibitors. Their interpretation in structural terms unveiled favorable substitutions toward selective CDK inhibitors and thus allowed for a rational design of specific ligands with minimized side effects. Additional 3D-quantitative structure-activity relationship (QSAR) analyses of a larger set of CDK-directed purines lead to the identification of essential structural requirements for affinity in this CDK ATP binding site. The combined interpretation of 3D-QSAR and the kinase target family landscape provides a consistent view to protein-ligand interactions, which are both favorable for affinity and selectivity in this important subfamily.

GRID/CPCA: a new computational tool to design selective ligands

We present a computational procedure aimed at understanding enzyme selectivity and guiding the design of drugs with respect to selectivity. It starts from a set of 3D structures of the target proteins characterized by the program GRID. In the multivariate description proposed, the variables are organized and scaled in a different way than previously published methodologies. Then, consensus principal component analysis (CPCA) is used to analyze the GRID descriptors, allowing the straightforward identification of possible modifications in the ligand to improve its selectivity toward a chosen target. As an important new feature the computational method is able to work with more than two target proteins and with several 3D structures for each protein. Additionally, the use of a 'cutout tool' allows to focus on the important regions around the active site. The method is validated for a total number of nine structures of the three homologous serine proteases thrombin, trypsin, and factor Xa. The regions identified by the method as being important for selectivity are in excellent agreement with available experimental data and inhibitor structure-activity relationships.

A 3D-QSAR study on the structural requirements for binding to CB(1) and CB(2) cannabinoid receptors

A 3D-QSAR study was carried out on 20 cannabinoids for which the binding affinities (K(i)) with respect to CB(1) and CB(2) receptors, determined in the same cell line, were available. For the first time three series of significantly different chemical structures such as Delta(9)-THC analogues, anandamides, and indoles were included in a single 3D-QSAR model, to obtain information on the interactions of all ligands with both CB(1) and CB(2) receptors and on their receptor selectivity. Delta(9)-THC was chosen as the structural template for alignment. The 3D-structure-activity correlation obtained by the GOLPE procedure provided a partial least squares (PLS) model with a very good predictive ability for the CB(1) receptor affinity of all compounds. The model allowed us to identify seven different regions in the space that contribute to explain the above binding affinities. External validation of the interpretation of the 3D-QSAR model was derived from a response-independent procedure such as principal components analysis (PCA). The CB(2) receptor model evidenced, besides the seven regions found for the CB(1) receptor, a new characteristic region for the CB(2) receptor. Another PCA, using 10 GRID probes, provided further evidence of receptor selectivity regions. One region opposite to the amidic NH of CB(1) selective O585 appears to be responsible for the CB(1) selectivity, while an interaction region opposite to the carbonyl of CB(2) selective JWH-015 appears to be involved in the CB(2) binding selectivity.

Chemometric rationalization of the structural and physicochemical basis for selective cyclooxygenase-2 inhibition: toward more specific ligands

The discovery that proinflammatory prostaglandins are produced by cyclooxygenase-2 (COX-2), an inducible isoform of the constitutive cyclooxygenase-1 (COX-1), opened a new frontier in the treatment of inflammatory diseases, because the selective inhibition of COX-2 can lead to therapeutically effective compounds which do not have the common side effects of classical non-steroidal antiinflammatory drugs (NSAIDs). Different crystallographic structures of both free COX-1 and COX-2 as well as complexes with inhibitors have been solved. Because of the great similarity between the two enzymes, it is difficult to detect the most important structural and physicochemical features that would be useful for designing inhibitors with an improved selectivity. In this paper we describe the application of a chemometric procedure to the study of COX-2 selective inhibition. This method, developed to reveal the most suitable regions of isoenzymes for the design of selective ligands, also has a very practical utility. GRID multivariate characterization of the enzymes and subsequent Principal Component Analysis (PCA) of the descriptor variables allow the identification of chemical groups that could be added to a core template structure to increase ligand selectivity.

Affinity and selectivity of matrix metalloproteinase inhibitors: a chemometrical study from the perspective of ligands and proteins

A novel strategy to understand affinity and selectivity for enzyme inhibitors using information from ligands and target protein 3D structures is described. It was applied to 2-arylsulfonyl-1,2,3, 4-tetrahydro-isoquinoline-3-carboxylates and -hydroxamates as inhibitors of the matrix metalloproteinases MMP-3 (stromelysin-1) and MMP-8 (human neutrophil collagenase). As the first step, consistent and predictive 3D-QSAR models were derived using CoMFA, CoMSIA, and GRID/Golpe approaches, leading to the identification of binding regions where steric, electronic, or hydrophobic effects are important for affinity. These models were validated using multiple analyses using two or five randomly chosen cross-validation groups and randomizations of biological activities. Second, 3D-QSAR models were derived based on the affinity ratio IC(50)(MMP-8)/IC(50)(MMP-3), allowing the identification of key ligand determinants for selectivity toward one of both enzymes. In addition to this ligands' view, the third step encompasses a chemometrical approach based on principal component analysis (PCA) of multivariate GRID descriptors to uncover the major differences between both protein binding sites with respect to their GRID probe interaction pattern. The resulting information, based on the accurate knowledge of the target protein 3D structures, led to a consistent picture in good agreement with experimentally observed differences in selectivity toward MMP-8 or MMP-3. The interpretation of all three classes of statistical models leads to detailed SAR information for MMP inhibitors, which is in agreement with available data for binding site topologies, ligand affinities, and selectivities. Thus the combined chemical analyses provide guidelines and accurate activity predictions for designing novel, selective MMP inhibitors.