Prediction of Drug-Drug Interaction Using an Attention-Based Graph Neural Network on Drug Molecular Graphs

The treatment of complex diseases by using multiple drugs has become popular. However, drug-drug interactions (DDI) may give rise to the risk of unanticipated adverse effects and even unknown toxicity. Therefore, for polypharmacy safety it is crucial to identify DDIs and explore their underlying mechanisms. The detection of DDI in the wet lab is expensive and time-consuming, due to the need for experimental research over a large volume of drug combinations. Although many computational methods have been developed to predict DDIs, most of these are incapable of predicting potential DDIs between drugs within the DDI network and new drugs from outside the DDI network. In addition, they are not designed to explore the underlying mechanisms of DDIs and lack interpretative capacity. Thus, here we propose a novel method of GNN-DDI to predict potential DDIs by constructing a five-layer graph attention network to identify k-hops low-dimensional feature representations for each drug from its chemical molecular graph, concatenating all identified features of each drug pair, and inputting them into a MLP predictor to obtain the final DDI prediction score. The experimental results demonstrate that our GNN-DDI is suitable for each of two DDI predicting scenarios, namely the potential DDIs among known drugs in the DDI network and those between drugs within the DDI network and new drugs from outside DDI network. The case study indicates that our method can explore the specific drug substructures that lead to the potential DDIs, which helps to improve interpretability and discover the underlying interaction mechanisms of drug pairs.     

Learning size-adaptive molecular substructures for explainable drug–drug interaction prediction by substructure-aware graph neural network

Drug–drug interactions (DDIs) can trigger unexpected pharmacological effects on the body, and the causal mechanisms are often unknown. Graph neural networks (GNNs) have been developed to better understand DDIs. However, identifying key substructures that contribute most to the DDI prediction is a challenge for GNNs. In this study, we presented a substructure-aware graph neural network, a message passing neural network equipped with a novel substructure attention mechanism and a substructure–substructure interaction module (SSIM) for DDI prediction (SA-DDI). Specifically, the substructure attention was designed to capture size- and shape-adaptive substructures based on the chemical intuition that the sizes and shapes are often irregular for functional groups in molecules. DDIs are fundamentally caused by chemical substructure interactions. Thus, the SSIM was used to model the substructure–substructure interactions by highlighting important substructures while de-emphasizing the minor ones for DDI prediction. We evaluated our approach in two real-world datasets and compared the proposed method with the state-of-the-art DDI prediction models. The SA-DDI surpassed other approaches on the two datasets. Moreover, the visual interpretation results showed that the SA-DDI was sensitive to the structure information of drugs and was able to detect the key substructures for DDIs. These advantages demonstrated that the proposed method improved the generalization and interpretation capability of DDI prediction modeling.

SA-DDI is designed to learn size-adaptive molecular substructures for drug–drug interaction prediction and can provide explanations that are consistent with pharmacologists.     

PASS-based prediction of metabolites detection in biological systems

ABSTRACT

Metabolite identification is an essential part of the drug discovery and development process. Experimental methods allow identifying metabolites and estimating their relative amount, but they require cost-intensive and time-consuming techniques. Computational methods for metabolite prediction are devoid of these shortcomings and may be applied at the early stage of drug discovery. In this study, we investigated the possibility of creating SAR models for the prediction of the qualitative metabolite yield (‘major’, ‘minor’, ”trace” and ”negligible”) depending on species and biological experimental systems. In addition, we have created models for prediction of xenobiotic excretion depending on its administration route for different species. The prediction is based on an algorithm of naïve Bayes classifier implemented in PASS software. The average accuracy of prediction was 0.91 for qualitative metabolite yield prediction and 0.89 for prediction of xenobiotic excretion. The created models were included as a component of MetaTox web application, which allows predicting the xenobiotic metabolism pathways (http://www.way2drug.com/mg).     

Computer-aided prediction for medicinal chemistry via the Internet1

Some computational tools for medicinal chemistry freely available on the Internet were compared to examine whether the results of prediction obtained with different methods coincided or not. It was shown that the correlation coefficients varied from 0.65 to 0.90 for log P (seven methods), from 0.01 to 0.73 for aqueous solubility (four methods), and from 0.19 to 0.73 for drug-likeness (three methods). While for log P estimates, reasonable average pairwise correlation was found, for aqueous solubility and drug-likeness it was rather poor. Therefore, using computational tools freely available via the Internet, medicinal chemists should evaluate their accuracy versus experimental data for particular series of compounds. In contrast to prediction of above mentioned properties, which can be done with several Internet tools, wide profiling of biological activity can be obtained only with PASS Inet (http://www.ibmc.msk.ru/PASS). PASS Inet was tested by a dozen medicinal chemists for compounds from different chemical series with various kinds of biological activity, and in the majority of cases the results of prediction coincided with the experiments. New anxiolytics, antiarrhythmics, antileishmanials, and other biologically active agents have been identified on this basis. The advantages and limitations of computer-aided predictions for medicinal chemistry via the Internet are discussed.     

Computer aided prediction of biological activity spectra: Evaluating versus known and predicting of new activities for thiazole derivatives

Computer aided prediction of biological activity spectra by the computer program PASS was applied to a set of 89 new thiazole derivatives. Experimentally tested activities (NSAID, local anaesthetic and antioxidant) coincide with the experiment in 70.8% cases, that exceeds significantly the random guess-work (～0.1%). Therefore, computer aided prediction using the Prediction of Activity Spectra for Substances (PASS) system (http://www.ibmh.msk.su/PASS) provides a reliable basis for planning of synthesis and experimental study for new compounds. New psychotropic activities are predicted for some compounds from the series under study. In particular, 7, 44 and 55 compounds likely have anxiolytic, anticonvulsant and cognition enhancer effects, respectively. Most of these compounds have the estimated values of probability to be active ( P a ) less than 60%. Therefore, if their activity will be confirmed by the experiment, they might occur to be New Chemical Entities.     

Polyfluoroaromatic compounds with a NCClR group—a reactive type of polyfluoroarene

This review deals with syntheses and reactivity of polyfluoroarenes with a NCClR group (including those with RCl). The most promising and convenient methods of synthesis of these type compounds are based on co-pyrolysis reactions of low-basic polyfluoroaromatic amines with RCCl₃ type derivatives and the interaction of such amines and polyfluoroaromatic hydrazines with RCX₃ (XCl,F) compounds in the presence of AlCl₃ at moderate temperature; they have no analogues in the nonfluorinated series. Reactions with different nucleophilic reagents, the formation of electrophilic intermediates of the nitrilium cation type and reactions of the latter with aromatic and fluoroaromatic hydrocarbons, amines, and compounds with CN, C≡N and CO multiple bonds are described. It is shown that polyfluoroaromatic compounds with a NCClR group are promising precursors for syntheses of a wide range of compounds of various classes, including heterocyclic derivatives.     

Primary amine diazeniumdiolate ions of structure {RNN(O)NOR} as ambident nucleophiles

An O²-substituted primary amine diazeniumdiolate RHN-N(O)NOR′ is ionized to an anion that attacks an electrophile R″X via either of two nitrogens to form both RR″N-N(O)NOR′ and R-NN(O)-N(R″)OR′. The bidentate character of such anions should be kept in mind when planning or evaluating drug discovery efforts involving diazeniumdiolates of structure RHN-N(O)NOR′.     

Tensile strength of ostrich carotid artery decellularized with liquefied dimethyl ether and DNase: An effort in addressing religious and cultural concerns

In a previous report, we proposed a method for decellularizing ostrich carotid arteries by removing lipids from the arteries using liquefied dimethyl ether (DME) instead of the conventional sodium dodecyl sulfate (SDS). This is followed by DNA fragmentation with DNase. In the present study, the physical properties of ostrich carotid arteries decellularized using the DME method were evaluated via tensile tests. Results showed that the tissue with SDS broke under less than half the stress of the original ostrich carotid artery. In contrast, the tissue treated with liquefied DME withstood the same level of the original tissue. Both decellularized tissues by liquefied DME showed flexibility comparable to that of the original tissue. We attributed this to the no CN bonds temporarily generated by dehydration through the Schiff-base reaction when lipids were removed by liquefied DME.     

Optimal designs for pairwise calculation: An application to free energy perturbation in minimizing prediction variability

Pairwise-based methods such as the free energy perturbation (FEP) method have been widely deployed to compute the binding free energy differences between two similar host–guest complexes. The calculated pairwise free energy difference is either directly adopted or transformed to absolute binding free energy for molecule rank ordering. We investigated, through both analytic derivations and simulations, how the selection of pairs in the experiment could impact the overall prediction precision. Our studies showed that (1) the estimated absolute binding free energy () derived from calculated pairwise differences (ΔΔG) through weighted least squares fitting is more precise in prediction than the pairwise difference values when the number of pairs is more than the number of ligands and (2) prediction precision is influenced by both the total number of pairs and the specifically selected pairs, the latter being critically important when the number of calculated pairs is limited. Furthermore, we applied optimal experimental design in pair selection and found that the optimally selected pairs can outperform randomly selected pairs in prediction precision. In an illustrative example, we showed that, upon weighing ligand structure similarity into design optimization, the weighted optimal designs are more efficient than the literature reported designs. This work provides a new approach to assess retrospective pairwise-based prediction results, and a method to design new prospective pairwise-based experiments for molecular lead optimization. © 2019 Wiley Periodicals, Inc.     

Luminescent rhenium(I)-gold(I) hetero organometallics linked by ethynylphenanthrolines

The first luminescent rhenium(I)-gold(I) hetero organometallics, Re{phenAu(PPh₃)}(CO)₃Cl (3) and Re{(PPh₃)AuphenAu(PPh₃)}(CO)₃Cl (4), have been prepared using the gold(I) complex AuCl(PPh₃) (PPh₃ = triphenylphosphine) and the novel rhenium(I) complexes Re(phenH)(CO)₃Cl (5) (phenH = 3-ethynyl-1,10-phenanthroline) or Re(HphenH)(CO)₃Cl (6) (HphenH = 3,8-bis(ethynyl)-1,10-phenanthroline). All the present rhenium(I) complexes 3-6 were revealed to possess a facial configuration (fac-isomer) with respect to the three carbonyl ligands. The main frameworks for these new gold(I) organometallics were constructed by the Au-C σ-bonding (with the η¹-type coordination) between the ethynylphenanthrolines and the Au(I) phosphine unit. Re(I)-Au(I) heterometallics 3 and 4 have shown single phosphorescence from the ³MLCT excited state and this observation can be interpreted in terms of the efficient intramolecular energy transfer from the Au(I) unit to the Re(I) unit.     

Prediction of biological activity spectra via the Internet

The majority of biologically active compounds have both pharmacotherapeutic and side/toxic actions. To estimate general efficacy and safety of the molecules under study, their biological potential should be thoroughly evaluated. In an early stage of study, only information about structural formulae was available and was used as an input for computational prediction. Based on a structural formulae of compounds presented as SDF or MOL-files, computer program PASS predicts 900 pharmacological effects, mechanism of action, and specific toxicity. An average accuracy of prediction in leave-one-out cross-validation is about 85%. For evaluating new compounds, scientific community may use PASS via the Internet for free at URL: http://www.ibmh.msk.su/PASS. In the first 18 months of PASS Inet's use, approximately 1000 researchers from 60 countries have obtained predicted biological activity spectra for about 23,000 different chemical compounds. More than 64 million PASS predictions for almost 250,000 compounds from Open NCI database are available on the web site http://cactus.nci.nih.gov/ncidb2/. These predictions are used for selecting compounds with desirable and without unwanted types of biological activities among the NCI samples available for screening.     

Twisted molecular geometry and localized electronic structure of the triplet excited gem-diphenyltrimethylenemethane biradical: substituent effects on thermoluminescence and related theoretical calculations

Substituent effects on the energies of electronic transitions (ETs) between the triplet excited and ground states of gem-diphenyltrimethylenemethane biradicals (³2a) were explored by using thermoluminescence (TL) spectroscopy and density functional theory (DFT) including time-dependent (TD) DFT. Linear free energy (Hammett) analyses of TL energies of a variety of para-substituted aryl derivatives of ³2* gave reasonable correlations with the substituent constant, σ. The slope of Hammett plots of the data are nearly identical to one obtained from a similar analysis of the photoluminescence (PL) energies of the structurally-related 1,1-diarylethyl radicals (3*). The results suggest that TL of ³2* and PL of 3* derive from a common diarylmethyl radical fluorophore. This interpretation is also supported by the DFT and TDDFT calculated electronic structures and ET energies of ³2 and 3. Thermodynamic and kinetic analyses of the charge recombination (CR) process between 2⁺ and 1⁻, which generates ³2*, revealed that substituents not only alter the TL energies but also the TL intensities of ³2*. The observations made in this effort demonstrate that ³2* as well as ³2 and 2⁺ have greatly twisted molecular geometries and highly localized electronic structures.     

Prediction of protein structural class using novel evolutionary collocation-based sequence representation

Knowledge of structural classes is useful in understanding of folding patterns in proteins. Although existing structural class prediction methods applied virtually all state-of-the-art classifiers, many of them use a relatively simple protein sequence representation that often includes amino acid (AA) composition. To this end, we propose a novel sequence representation that incorporates evolutionary information encoded using PSI-BLAST profile-based collocation of AA pairs. We used six benchmark datasets and five representative classifiers to quantify and compare the quality of the structural class prediction with the proposed representation. The best, classifier support vector machine achieved 61-96% accuracy on the six datasets. These predictions were comprehensively compared with a wide range of recently proposed methods for prediction of structural classes. Our comprehensive comparison shows superiority of the proposed representation, which results in error rate reductions that range between 14% and 26% when compared with predictions of the best-performing, previously published classifiers on the considered datasets. The study also shows that, for the benchmark dataset that includes sequences characterized by low identity (i.e., 25%, 30%, and 40%), the prediction accuracies are 20-35% lower than for the other three datasets that include sequences with a higher degree of similarity. In conclusion, the proposed representation is shown to substantially improve the accuracy of the structural class prediction. A web server that implements the presented prediction method is freely available at http://biomine.ece.ualberta.ca/Structural_Class/SCEC.html.     

Synthesis and characterization of potential impurities of Vigabatrin-An anti epileptic drug

The present work describes the synthesis and characterization of four potential impurities of Vigabatrin (1 Grant, S. M.; Heel, R. C. A Review of Its Pharmacodynamic and Pharmacokinetic Properties, and Therapeutic Potential in Epilepsy and Disorders of Motor Control. Drugs. 1991, 41, 889-926. doi:10.2165/00003495-199141060-00007[Crossref], [PubMed], [Web of Science ®] , [Google Scholar]) namely 2-(2-aminobut-3-enyl)malonic acid (2 Impurities in New Drug Substances Q3A (R2): 2006. www.ich.org/products/guidelines/quality/article/quality-guidelines.html (accessed Aug 16, 2018). [Google Scholar]) (Vigabatrin USP impurity-E), 2-(2-oxo-5-vinylpyrrolidin-1-yl)acetic acid (3 Validation of Analytical Procedures: Text and Methodology Q2 (R1): 2006. www.ich.org/products/guidelines/quality/article/quality-guidelines.html (accessed Aug 16, 2018). [Google Scholar]) (USP Tablets impurity), 4-aminohexanoic acid (4 Maurice, W. G.; Gerard, J. L. 2-Pyrrolidinone Compounds and Processes for Making Same. U.S. Patent 4,235,778, November 25, 1980. [Google Scholar]) and 2,2′-oxo-5,5′-bispyrrolidinyl ether (5 Marie-Christine, D.; Bienayme, H.; Popowycz, F.; Lemaire M. Process for Preparing 4-Amino-5-hexenoic acid From Succinimide. E.P. Patent 2,537,827 A1, June 26, 2011. [Google Scholar]). Compound 4 is a possible process related impurity of 1 where as compound 5 is a process related impurity of 5-ethoxy-2-pyrrolinone (16). All these impurities have a significant impact on the quality of the drug product. This work is extremely useful for generic pharmaceutical industry.     

Organometallic chemistry and catalysis on gold metal surfaces

As in transition metal complexes, CN-R ligands adsorbed on powdered gold undergo attack by amines to give putative diaminocarbene groups on the gold surface. This reaction forms the basis for the discovery of a gold metal-catalyzed reaction of CN-R, primary amines (R′NH₂) and O₂ to give carbodiimides (R′-NCN-R). An analogous reaction of CO, RNH₂, and O₂ gives isocyanates (R-NCO), which react with additional amine to give urea (RNH)₂CO products. The gold-catalyzed reaction of CN-R with secondary amines (HNR′₂) and O₂ gives mixed ureas RNH(CO)NR′₂. In another type of gold-catalyzed reaction, secondary amines HN(CH₂R)₂ react with O₂ to undergo dehydrogenation to the imine product, RCHN(CH₂R). Of special interest is the high catalytic activity of gold powder, which is otherwise well-known for its poor catalytic properties.     

Further chemistry of ruthenium butatrienylidene complexes

Several complexes have been obtained from reactions carried out in early attempts to prepare the diynyl complexes Ru(CCCCR)(dppe)Cp* (R = H, SiMe₃). These have been identified crystallographically as the acyl complex Ru{CCC(O)Me}(dppe)Cp* (3), the cationic imido complex [Ru{CCC(NH₂)Me}(dppe)Cp*]PF₆ (4), the binuclear butenynylallenylidene [{Ru(dppe)Cp*}₂{μ-CCC(OMe)CHCMeCC}]PF₆ (5), and the bis(ethynyl)cyclobutenylidene [{Ru(dppe)Cp*}₂{μ-CCC₄H₂(SiMe₃)CC}]PF₆ (6). NMR studies of 5 have revealed the existence of two isomers. Plausible routes for their formation from the putative butatrienylidene intermediate [Ru(CCCCH₂)(dppe)Cp*]⁺ (A) are discussed.     

Further reactions of some bis(vinylidene)diruthenium complexes

Whereas {Ru(dppm)Cp*}₂(μ-CCCC) (2) is the only product formed by deprotonation of [{Ru(dppm)Cp*}₂{μ(CCHCHC)}]⁺ with dbu, a mixture of 2 with Ru{CCCHCH(PPh₂)₂[RuCp*]}(dppm)Cp* (3) and {Cp*Ru(PPh₂CHCCH-)}₂ (4) is obtained with KOBu^t. A similar reaction with [{Ru(dppm)Cp*}₂{μ(CCMeCMeC)}]⁺ (5) gave Ru{CCCMeCH(PPh₂)₂[RuCp*]}(dppm)Cp* (6). X-ray structures of 4, 5 and 6 confirm the presence of the 1-ruthena-2,4-diphosphabicyclo[1.1.1]pentane moiety, which is likely formed by an intramolecular attack of the deprotonated dppm ligand on C(1) of the vinylidene ligand. Protonation of {Ru(dppe)Cp*}₂(μ-CCCC) (8-Ru) regenerates its precursor [{Ru(dppe)Cp*}₂{μ(CCHCHC)}]²⁺ (7-Ru). Ready oxidation of the bis(vinylidene) complex affords the cationic carbonyl [Ru(CO)(dppe)Cp*]PF₆ (9) (X-ray structure).