Current Computer - Aided Drug Design - Volume 12, Issue 1, 2016

Compound selectivity prediction plays an important role in identifying potential compounds that bind to the target of interest with high affinity. However, there is still short of efficient and accurate computational approaches to analyze and predict compound selectivity. In this paper, we propose two methods to improve the compound selectivity prediction. We employ an improved multitask learning method in Neural Networks (NNs), which not only incorporates both activity and selectivity for other targets, but also uses a probabilistic classifier with a logistic regression. We further improve the compound selectivity prediction by using the multitask learning method in Deep Belief Networks (DBNs) which can build a distributed representation model and improve the generalization of the shared tasks. In addition, we assign different weights to the auxiliary tasks that are related to the primary selectivity prediction task. In contrast to other related work, our methods greatly improve the accuracy of the compound selectivity prediction, in particular, using the multitask learning in DBNs with modified weights obtains the best performance.

Inverse (or reverse) docking approach which involves docking of a ligand against a set of protein structures to predict possible protein target(s), possess limitations, including inefficient empirical scoring schemes and similarities in protein active site shape and physico-chemical properties. To overcome this limitation, we combined receptor- and ligand-based methods to predict probable protein targets. We showed that the experimental protein target along with possible offtargets can be effectively retrieved if the docking energy of the reference molecule and probe molecules based scaled energy profiles were combined and clustered together. The present method was validated using 7,8-dialkyl-1,3-diaminopyrrolo[3,2-f]quinazolines that inhibit Candida albicans dihydrofolate reductase (DHFR) in vitro.

The increasing mortality due to antibacterial resistance necessitates the search for novel antimicrobial agents. Hence, series of 1-R-2-([1,2,4]triazolo[1,5-c]quinazoline-2-ylthio)etanon(ol)s were synthesized, evaluated by spectral data and studied against St. aureus, M. luteum, E. faecalis, E. aerogenes, P. aeruginosa, C. sakazakii, E. coli, K. pneumonia, hospital Streptococcus spp., C. albicans and A. niger in 100, 500 μg/mL and 100 μg/disk. Substances exhibited moderate toxicity in 0.025, 0.1 and 0.25 mg/mL in bioluminescence inhibition tests of Photobacterium leiognathi. SAR exposed that introduction of 2,4-(Cl)2C6H3-, 2,5-(OMe)2C6H3-, 4-Me-2-iPr-C6H3O- and 3-iPr-C6H4O- fragments and reduction of the pyrimidine ring of R-([1,2,4]triazolo[1,5-c]quinazolin-2-ylthio)alcohols were the best modifications to promote antimicrobial activity. Molecular docking showed their good affinity into the active sites of EcPanK-AMPPNP and hDHFR. Hence, reported results will be used for subsequent QSAR model creation and purposeful antimicrobial modification of the strongest compounds.

Background: Human immunodeficiency virus type 1 (HIV-1) infection ultimately leading to acquired immunodeficiency syndrome (AIDS), remains a significant problem. CCR5 is a member of the chemokine receptor family that is utilized in the early stage of the replication cycle by the most commonly transmitted M-tropic strains of HIV-1. In this study, we developed 3D-QSAR models using CoMFA and CoMSIA methods on a series of 71 imidazolidinylpiperidinylbenzoic acid CCR5 antagonists, in order to better understand the substituent requirements and get more potent antagonists of CCR5. Methods: The research of 3D-QSAR modeling of imidazolidinylpiperidinylbenzoic acids as chemokine receptor 5 (CCR5) antagonists was conducted using comparative molecular field analysis (CoMFA) and comparative molecular similarity index analysis (CoMSIA). Results: For this study, a dataset containing 71 imidazolidinyl-piperidinyl-benzoic acids was divided into a training set of 22 compounds and a test set of 49 compounds. The results obtained from the CoMFA/CoMSIA model exhibited a statistical significance r2 of 0.996 (0.984) with an estimated standard error of 0.109 (0.209). Conclusion: Both CoMFA and CoMSIA models provided valuable insight into the structural requirements for improving the activity of then CCR5 antagonists.

Poor pharmacokinetic and toxicity profiles are major reasons for the low rate of advancing lead drug candidates into efficacy studies. The In-silico prediction of primary pharmacokinetic and toxicity properties in the drug discovery and development process can be used as guidance in the design of candidates. In-silico parameters can also be used to choose suitable compounds for in-vivo testing thereby reducing the number of animals used in experiments. At the Novartis Institute for Tropical Diseases, a data set has been curated from in-house measurements in the disease areas of Dengue, Tuberculosis and Malaria. Volume of distribution, half-life, total in-vivo clearance, in-vitro human plasma protein binding and in-vivo oral bioavailability have been measured for molecules in the lead optimization stage in each of these three disease areas. Data for the inhibition of the hERG channel using the radio ligand binding dofetilide assay was determined for a set of 300 molecules in these therapeutic areas. Based on this data, Artificial Neural Networks were used to construct In-silico models for each of the properties listed above that can be used to prioritize candidates for lead optimization and to assist in selecting promising molecules for in-vivo pharmacokinetic studies.

We have carried out systematic large-basis set quantum chemical computations at Møller- Plesset second-order perturbation (MP2) and couple cluster singles + doubles CCSD and CCSD(T)with triples correction levels of theories on a set of 55 halogenated carbons in the Crebelli toxicological dataset. We have computed a number of electronic properties at optimized geometries such as vertical electron affinities, HOMO-LUMO gaps, dipole moments, etc. We have provided insights into the mechanism of toxicity through electron attachment in metabolic pathways by binding to an electron donating enzyme in hepatocytes. The electron transfer from the enzyme to the halocarbon is accompanied by bond elongation resulting in autodetachment as evidenced from potential energy surfaces of the anion and neutral molecule. The autodetachment process leads to production of highly reactive free radicals, which cause tissue damage, and prolonged exposure can result in hepatocellular carcinoma depending on the hydrogen extraction propensity of the free radical and vertical electron affinity of the neutral halocarbon.

The protease β-secretase plays a critical role in the synthesis of pathogenic amyloid-β in Alzheimer’s disease. In this study, pharmacophore constructed from receptor-ligand complex was used to screen Chemdiv and Zinc database and the resulting hits were subjected to docking experiments using LiandFit and CDOCKER programs. Molecules with high consensus scores and good interaction patterns in docking programs were retained. Drug-likeness assay including Lipinski’s rule of five and ADMET properties filters were further used to identify BACE1 inhibitor. Finally, 13 compounds with novel scaffolds were selected and, considering of the nature of relative high LogP value of many marketed AD drugs, three of them with top 3 predicted LogP value were evaluated for their IC50 values in vitro by BACE1 enzymatic activity study. We believe that compound 13 with an IC50 value of 136 μM can be a lead compound with great potential in BACE1 inhibition and increasing activity by subsequently structure modification or optimization. At the same time, we found that the interaction between the residues Asp228, Asp32 of BACE1 and ligands is significant through analyzing the binding mode of 13 candidate compounds.