GRID Molecular Interaction Fields Research Articles

DeepGRID: Deep Learning Using GRID Descriptors for BBB Prediction.

Deep Learning approaches are able to automatically extract relevant features from the input data and capture nonlinear relationships between the input and output. In this work, we present the GRID-derived AI (GrAId) descriptors, a simple modification to GRID MIFs that facilitate their use in combination with Convolutional Neural Networks (CNNs) to build Deep Learning models in a rotationally, conformationally, and alignment-independent approach we are calling DeepGRID. To our knowledge, this is the first time that GRID MIFs have been combined with CNNs in a Deep Learning approach. We applied the approach to build regression and classification models for blood-brain barrier permeation, an important factor when designing CNS drugs and conversely when designing to avoid off-target effects for CNS-inactive drugs. The VolSurf approach was one of the first to successfully model this property from three-dimensional structures, using descriptors derived from their GRID Molecular Interaction Fields (MIFs) in combination with PLS. We compared the DeepGRID models with others built using the hand-crafted VolSurf descriptors in combination with both PLS and Random Forest (RF). Both the DeepGRID and RF regression models performed best according to the % of compounds with a Geometric Mean Fold Error (GMFE) within 2-fold of the experimental data. Applying these regression models as classifiers, for the smaller 332 and 416 compound data sets all models performed well with ROC AUC values of ∼0.9 on the external test set. For the larger 2105 compound data set, the DeepGRID classifier performed the best with an AUC of 0.87 on the external test set with the RF model having an AUC of 0.84 and the original VolSurf lgBB model having an AUC of 0.83.

FragExplorer: GRID-Based Fragment Growing and Replacement.

Understanding which chemical modifications can be made to known ligands is a key aspect of structure-based drug design and one that was pioneered by the software GRID. We developed FragExplorer with the explicit aim of showing GRID users which fragments would best match the GRID molecular interaction fields in a protein binding site, given a bound ligand as a starting point. Users can grow ligands or replace existing moieties; the R-Group Exploration mode identifies all potential R-Groups and searches for replacements automatically; the Scaffold Exploration mode does the same for all potential scaffolds. For a ligand with three points of variation, R-Group Exploration will typically explore a chemical space of 1016 potential molecules; including Scaffold Exploration increases this to 1022. FragExplorer was designed to be integrated within an interactive 3D Editor/Designer; therefore, the speed of computation was an important consideration; a typical fragment search takes 20 seconds. In a fragment reprediction test, FragExplorer demonstrates an overall fragment retrieval rate of 55%, increasing to 69% for smaller fragments. At a 90% substructural match, the retrieval rate increases to ∼80%. We also show how the approach could have been used to hop from olmesartan to azilsartan or to optimize a p38 MAP kinase lead to a compound that bears similarity to a known nanomolar inhibitor.

Molecular modelling of epitopes recognized by neoplastic B lymphocytes in Chronic Lymphocytic Leukemia

Identification of epitopes recognized by tumour B cells could provide insights into the molecular mechanisms of B cell tumorigenesis through aberrant B cell receptor (BCR) signalling. Here, we analysed the structure of eleven peptides binders of BCRs expressed in Chronic Lymphocytic Leukemia (CLL) patients in order to identify the chemical features required for cross-reactive binding to different CLL clonotypes. Four cross-reactive (CR) and seven no-cross-reactive (NCR) peptides were analysed by means of GRID molecular interaction fields, ligand-based pharmacophore and 3D-QSAR approaches. Based on pharmacophore model, two peptides were generated by specific amino acids substitutions of the parental NCR peptides; these new peptides resumed the common chemical features of CR peptides and bound the CLL BCR clonotypes recognized by CR peptides and parental NCR peptides. Thus, our computational approach guided the pharmacophore modelling of CR peptides. In perspective, peptide binders of CLL BCR clonotypes could represent a powerful tool for computational modelling of epitopes recognized by tumour B cells clones.

Non-Zinc-Binding Inhibitors of MMP-13: GRID-Based Approaches to Rationalize the Binding Process.

Matrix metalloproteinases (MMPs) are zinc enzymes responsible for the degradation of the extracellular matrix. With this function, MMPs are involved in many physiological processes, but also in many pathological states. MMP-13 is implicated in the degradation of type II collagen, the main structural protein of articular cartilage, contributing to the development of osteoarthritis and inflammatory diseases. In the last years, a new generation of potent and selective MMP inhibitors (MMPIs) has been identified and classified as non-zinc-binding inhibitors (NZBIs). Several MMP-13 NZBIs have been developed and crystallographically determined in complex with the enzyme. Here, we provide a detailed review of the current knowledge about this class of MMP-13 inhibitors and, by using computational procedures, we highlight the molecular determinants that are needed for the binding process. In particular, FLAP, a program based on GRID molecular interaction fields, was used to analyze the ligand-protein interactions: molecular shape and hydrogen bond acceptor groups strongly influence the binding according to the ligand-based modeling, while the aromatic interactions are better identified by the structure-based study. The complementary results can be combined in a high performance model, showing the effectiveness of molecular interaction field based approaches to search for novel MMP-13 NZBIs.

3D-QSAR studies and pharmacophore identification of AT1 receptor antagonists

A 3D-QSAR model using the GRIND/ALMOND descriptors has been performed on a set of 49 angiotensin receptor blockers, also known as angiotensin II receptor antagonists, a family of agents that bind to and inhibit the angiotensin II type 1 receptor. The most commonly used chemical probes: DRY (hydrophobic interaction), O (carbonyl oxygen sp2, hydrogen bond donor), N1 (NH neutral, hydrogen bond acceptor) and TIP (shape descriptor molecular forms) were derived from the GRID molecular interaction fields. A statistical approach was undertaken using the method of partial least squares within the Pentacle program. The results show satisfactory accuracy of the prediction model (RMSEE = 0.239, R2 = 0.94, Q2 = 0.85). The V597 (DRY-TIP) and V763 (O-TIP) represent the most significant variables that correlate positively with the activity of the ARBs. Thirty novel structures of ARBs were designed, according to the developed 3D-QSAR model and pharmacophore, what might set the basis for development of new antihypertensive drugs.

An Alignment Independent 3D-QSAR Modeling of Dispersibility of Single-walled Carbon Nanotubes in Different Organic Solvents

An alignment free, three dimensional quantitative structure activity relationships (3D-QSAR) of dispersibility of single walled carbon nanotubes (SWNTs) in a diverse set of organic solvents was reported for the first time. GRIND methodology, where descriptors are derived from GRID molecular interaction fields (MIF), was used. Different variable selection procedures including: fractional factorial design (FFD), stepwise multiple linear regression (SW-MLR), successive projection algorithm (SPA), genetic algorithm (GA), and enhanced replacement method (ERM) were used to extract the more informative factors from exported GRIND descriptors and generate more predictive model. Partial least square (PLS) was applied to model construction and ERM-PLS based GRIND descriptors showed excellent performance in predicting of SWNTs dispersibility. ERM-PLS model satisfied a set of rigorous validation criteria and performed well in the prediction of an external test set. From the GRIND variables involved in ERM-PLS model the identification of some key molecular features and their position in solvent structure, which is crucial in SWNTs disperibility, would be possible. The obtained results confirmed the importance of hydrophobic interactions, size and steric hindrance of hydrophibic part of solvent molecule. Interestingly, the effect of presence of a hydrogen bond donor or polar group in structure of a solvent molecule with a large size couldn't be neglected.

GRID-Based Three-Dimensional Pharmacophores I: FLAPpharm, a Novel Approach for Pharmacophore Elucidation

Pharmacophore elucidation approaches are routinely used in drug discovery, primarily with the aim of determining the three-dimensional arrangement of common features shared by ligands interacting at the site of interest; these features can then be used to investigate the structure-activity relationship between the ligands and also to screen for other molecules possessing the relevant features. Here we present a novel approach based on GRID molecular interaction fields and the derivative method FLAP that has been previously described, which provides a common reference framework to compare both small molecule ligands and macromolecular protein targets. Unlike classical pharmacophore elucidation approaches that extract simplistic molecular features, determine those which are common across the data set, and use these features to align the structures, FLAPpharm first aligns the structures and subsequently extracts the common interacting features in terms of their molecular interaction fields, pseudofields, and atomic points, representing the common pharmacophore as a more comprehensive pharmacophoric pseudomolecule. The approach is applied to a number of data sets to investigate performance in terms of reproducing the X-ray crystallography-based alignment, in terms of its discriminatory ability when applied to virtual screening and also to illustrate its ability to explain alternative binding modes. In part two of this publication, a comprehensive benchmark data set for pharmacophore elucidation is presented and the performance of FLAPpharm discussed.

An alignment independent 3D-QSAR study for predicting the stability constants of structurally diverse compounds with β-cyclodextrin

For the first time, to our opinion, an alignment free, three dimensional quantitative structure activity relationships (3D-QSAR) of stability constants of a large and heterogeneous variety of organic substances with β-cyclodextrin was reported. GRIND methodology, where descriptors are derived from GRID molecular interaction fields (MIF), was used. After variable selection via fractional factorial design (FFD), PLS analysis was carried out, and a highly descriptive and predictive model was obtained. The model satisfied a set of rigorous validation criteria and performed well in the prediction of an external test set. The proposed model is also checked for free from chance correlation, reliability and robustness by permutation testing called progressive y scrambling. The obtained models confirmed that size and shape of the molecules as well as hydrophobic interactions are the main parameters influencing the stabilities of diverse compounds—β-cyclodextrin inclusion complexes.

An alignment independent 3D QSAR study of the antiproliferative activity of 1,2,4,5-tetraoxanes

An alignment-free 3D QSAR study on antiproliferative activity of the thirty-three 1,2,4,5-tetraoxane derivatives toward two human dedifferentiated cell lines was reported. GRIND methodology, where descriptors are derived from GRID molecular interaction fields (MIF), were used. It was found that pharmacophoric pattern attributed to the most potent derivatives include amido NH of the primary or secondary amide, and the acetoxy fragments at positions 7 and 12 of steroid core which are, along with the tetraoxane ring, common for all studied compounds. Independently, simple multiple regression model obtained by using the whole-molecular properties, confirmed that the hydrophobicity and the H-bond donor properties are the main parameters influencing potency of compounds toward human cervix carcinoma (HeLa) and human malignant melanoma (FemX) cell lines. Corollary, similar structural motifs are found to be important for the potency toward both examined cell lines.

Antiproliferative activity of NCI-DTP glutarimide derivatives. An alignment independent 3D QSAR study

Alignment-free, three dimensional structure-activity relationships (3D QSAR) of the antiproliferative potency of twenty-two glutarimide-containing compounds, taken from National Cancer Institute Developmental therapeutics Program database, toward eight representative human tumour cell lines are reported. The descriptors used in the QSAR study were derived from GRID molecular interaction fields. The obtained models readily detect structural motifs positively or negatively correlated with the potency of the studied compounds toward each cell line. In this way, the pharmacophoric pattern required for high potency of compounds is reported. This pattern can serve as guidance for the design and syntheses of novel congeners, planed to be tested toward human tumour cell lines.

High-Throughput Virtual Screening of Proteins Using GRID Molecular Interaction Fields

A new computational algorithm for protein binding sites characterization and comparison has been developed, which uses a common reference framework of the projected ligand-space four-point pharmacophore fingerprints, includes cavity shape, and can be used with diverse proteins as no structural alignment is required. Protein binding sites are first described using GRID molecular interaction fields (GRID-MIFs), and the FLAP (fingerprints for ligands and proteins) method is then used to encode and compare this information. The discriminating power of the algorithm and its applicability for large-scale protein analysis was validated by analyzing various scenarios: clustering of kinase protein families in a relevant manner, predicting ligand activity across related targets, and protein-protein virtual screening. In all cases the results showed the effectiveness of the GRID-FLAP method and its potential use in applications such as identifying selectivity targets and tools/hits for new targets via the identification of other proteins with pharmacophorically similar binding sites.

SHOP: A Method For Structure‐Based Fragment and Scaffold Hopping

A new method for fragment and scaffold replacement is presented that generates new families of compounds with biological activity, using GRID molecular interaction fields (MIFs) and the crystal structure of the targets. In contrast to virtual screening strategies, this methodology aims only to replace a fragment of the original molecule, maintaining the other structural elements that are known or suspected to have a critical role in ligand binding. First, we report a validation of the method, recovering up to 95% of the original fragments searched among the top-five proposed solutions, using 164 fragment queries from 11 diverse targets. Second, six key customizable parameters are investigated, concluding that filtering the receptor MIF using the co-crystallized ligand atom type has the greatest impact on the ranking of the proposed solutions. Finally, 11 examples using more realistic scenarios have been performed; diverse chemotypes are returned, including some that are similar to compounds that are known to bind to similar targets.

Prediction of the binding mode of biarylpropylsulfonamide allosteric AMPA receptor modulators based on docking, GRID molecular interaction fields and 3D-QSAR analysis

A novel approach of combining flexible molecular docking, GRID molecular interaction fields, analysis of ligand–protein hydrogen bond interactions, conformational energy penalties and 3D-QSAR analysis was used to propose a binding mode in the dimer interface of the iGluR2 receptor for the biarylpropylsulfonamide class of positive allosteric AMPA modulators. Possible binding poses were generated by flexible molecular docking. GRID molecular interaction fields of the binding site, ligand–protein hydrogen bonding interactions and conformational energy penalties were used to select the most likely binding mode. The selected binding poses were subjected to a 3D-QSAR analysis using previously published activity data. The resulting model (2 LVs, R 2 = 0.89, q 2 = 0.61) predicted the activities of the compounds in the test set with a standard deviation on error of prediction of 0.17. The proposed binding mode was validated by interpretation of the PLS-coefficient regions from the 3D-QSAR analysis in terms of interactions between the receptor and the modulators.

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.

Chemometric Characterization of Chromatographic Retention Parameters of Mesoionic 1,3,4‐Thiadiazolium‐3‐Aminides by Molecular Interaction Fields

Mesoionic 1,3,4‐thiadiazolium‐3‐aminides are described according to their properties of chromatographic partitioning. Log values obtained from C18 and in‐house poly(methyloctylsiloxane) chromatographic columns correlated well with log P heteroatom contribution to overall lipophilicity. The pivotal role of partitioning from mobile to stationary phases is unraveled with GRID Molecular Interaction Fields, MIF. The descriptors extracted by the program VolSurf, resulted in partitioning being dependent on the hydrophilic‐lipophilic balance. Volume‐related terms and electrostatics of the heterocyclic ring are of major importance in the retention mechanism.

Virtual Screening and Scaffold Hopping Based on GRID Molecular Interaction Fields

In this study, a set of strategies for structure-based design using GRID molecular interaction fields (MIFs) to derive a pharmacophoric representation of a protein is reported. Thrombin, one of the key enzymes involved in the blood coagulation cascade, was chosen as the model system since abundant published experimental data are available related to both crystal structures and structurally diverse sets of inhibitors. First, a virtual screening methodology was developed either using a pharmacophore representation of the protein based on GRID MIFs or using GRID MIFs from the 3D structure of a set of chosen thrombin inhibitors. The search was done in a 3D multiconformation version of the Available Chemical Directory (ACD) database, which had been spiked with 262 known thrombin inhibitors (multiple conformers available per compound). The model managed to find 80% of the known thrombin inhibitors among the 74,291 conformers in the ACD by only searching 5% of the database; hence, a 15-fold enrichment of the library was achieved. Second, a scaffold hopping methodology was developed using GRID MIFs, giving the scaffold interaction pattern and the shape of the scaffold, together with the distance between the anchor points. The scaffolds reported by Dolle in the Journal of Combinatorial Chemistry summaries (2000 and 2001) and scaffolds built or derived from ligands cocomplexed with the thrombin enzyme were parameterized using a new set of descriptors and saved into a searchable database. The scaffold representation from the database was then compared to a template scaffold (from a thrombin crystal structure), and the thrombin-derived scaffolds included in the database were found among the top solutions. To validate the usefulness of the methodology to replace the template scaffold, the entire molecule was built (scaffold and side chains) and the resulting compounds were docked into the active site of thrombin. The docking solutions showed the same binding pattern as the cocomplexed compound, hence, showing that this method can be a valuable tool for medicinal chemists to select interchangeable core structures (scaffolds) in an easy manner and retaining the binding properties from the original ligand.

Predicting drug metabolism: a site of metabolism prediction tool applied to the cytochrome P450 2C9.

The aim of the present study is to develop a method for predicting the site at which molecules will be metabolized by CYP 2C9 (cytochrome P450 2C9) using a previously reported protein homology model of the enzyme. Such a method would be of great help in designing new compounds with a better pharmacokinetic profile, or in designing prodrugs where the compound needs to be metabolized in order to become active. The methodology is based on a comparison between alignment-independent descriptors derived from GRID Molecular Interaction Fields for the CYP 2C9 active site, and a distance-based representation of the substrate. The predicted site of metabolism is reported as a ranking list of all the hydrogen atoms of each substrate molecule. Eighty-seven CYP 2C9-catalyzed oxidative reactions reported in the literature have been analyzed. In more than 90% of these cases, the hydrogen atom ranked at the first, second, or third position was the experimentally reported site of oxidation.

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.