Accumulated evidence suggests that the in vivo biological potency of a ligand is more strongly correlated with the binding/unbinding kinetics than the equilibrium thermodynamics of the protein–ligand interaction (PLI). However, the existing experimental and computational techniques are largely insufficient and limited in large-scale measurements or accurate predictions of the kinetic properties of PLI. In this work, elaborate efforts have been made to develop interconsistent, reasonable, and predictive models of the association rate constant (kon), dissociation rate constant (koff), and equilibrium dissociation constant (KD) of a series of HIV protease inhibitors with different structural skeletons. The results showed that nine Volsurf descriptors derived from water (OH2) and hydrophobic (DRY) probes are key molecular determinants for the kinetic and thermodynamic properties of HIV-1 protease inhibitors. To the best of our knowledge, this is the first time that interconsistent and reasonable models with strong prediction power have been established for both the kinetic and thermodynamic properties of HIV protease inhibitors.

CYP1A2, an important member of the cytochromes P450 (CYPs) superfamily, is involved in the metabolism or bioactivation of many clinical drugs and precarcinogens. Thus, accurate prediction of CYP1A2 inhibitors is of great importance in early drug discovery and cancer prevention. In this study, a dataset of more than 12000 structurally diverse compounds was used to develop prediction models by a support vector machine (SVM). By combining two types of fragment descriptors, i.e. Molecular Hologram and MACCS descriptors, an improved radial basis function (RBF)-based SVM model was obtained, of which the accuracies (ACCs), sensitivities (SENs), specificities (SPEs), and Matthews correlation coefficients (MCCs) were 90.95%, 92.40%, 89.70%, 0.8191 for 6396 training samples, and 83.14%, 85.17%, 81.41%, 0.6638 for 6395 test samples, respectively. The prediction capability of the SVM model obtained was further validated by an independent dataset of 2581 samples with geometric mean (G-mean) based accuracy of 70.67%. The results indicate that the combination of the two types of fragment descriptors is an extremely efficient method for eliciting the key structural features of CYP inhibitors, and thus can be employed to large-scale virtual screening of inhibitors of CYP isoforms.

Flexible peptides binding to human leukocyte antigen (HLA) play a key role in mediating human immune responses and are also involved in idiosyncratic adverse drug reactions according to recent research. However, the structural determinations of pHLA complexes remain challenging under the present conditions. In this paper, the performance of a new peptide docking method, namely FlexPepDock, was systematically investigated by a benchmark of 30 crystallized structures of peptide-HLA class I complexes. The docking results showed that the near-native pHLA-I models with peptide bb-RMSD less than 2 Å were ranked in the top 1 model for 100% (70/70) docking cases, and the subangstrom models with peptide bb-RMSD less than 1 Å were ranked in the top 5 lowest-energy models for 65.7% (46/70) docking cases. Furthermore, 10 out of 70 docking cases ranked the subangstrom all-atom models in the top 5 lowest-energy models. The results showed that the FlexPepDock can generate high-quality models of pHLA-I complexes and can be widely applied to pHLA-I modeling and mechanism research of peptide-mediated immune responses.

•Data credibility analysis is used for selecting training samples.•A predictive ECP model is obtained with sensitivities larger than 0.95.•ECP is attractive for virtual screening when few positive samples are available.Recently, emerging chemical patterns (ECPs) has been proposed as a powerful tool for compound classification in cheminformatics. However, the prediction power and applicability of the ECP approach has remained largely unexplored. Herein, the effects of sample size, data quality, and unbalanced data on the prediction performance of ECP were systematically investigated by using a dataset consisted of 666 P-gp inhibitors and 609 non-inhibitors. The results showed that the ECP classification can achieve high sensitivity and modest specificities, depending on the size or positive-to-negative ratio of a training set. For a training set with only 3 positive and 3 negative training samples, a predictive ECP model was obtained with sensitivity larger than 0.95 for 418 test samples. In addition, the results showed that the prediction performance of an ECP model was strongly influenced by the quality of training samples. Taken together, the ECP approach renders methodology attractive for the virtual screening of lead compounds, especially when few positive samples are available.

P-glycoprotein (P-gp) is an ATP-binding cassette multidrug transporter. The over expression of P-gp leads to the development of multidrug resistance (MDR), which is a major obstacle to effective treatment of cancer. Thus, designing effective P-gp inhibitors has an extremely important role in the overcoming MDR. In this paper, both ligand-based quantitative structure–activity relationship (QSAR) and receptor-based molecular docking are used to predict P-gp inhibitors. The results show that each method achieves good prediction performance. According to the results of tenfold cross-validation, an optimal linear SVM model with only three descriptors is established on 857 training samples, of which the overall accuracy (Acc), sensitivity, specificity, and Matthews correlation coefficient are 0.840, 0.873, 0.813, and 0.683, respectively. The SVM model is further validated by 418 test samples with the overall Acc of 0.868. Based on a homology model of human P-gp established, Surflex-dock is also performed to give binding free energy-based evaluations with the overall accuracies of 0.823 for the test set. Furthermore, a consensus evaluation is also performed by using these two methods. Both QSAR and molecular docking studies indicate that molecular volume, hydrophobicity and aromaticity are three dominant factors influencing the inhibitory activities.

Recently, genome wide association studies showed that there is a strong association between abacavir-induced serious, idiosyncratic, adverse drug reactions (ADRs) and human leukocyte antigen-B*5701 (HLA-B*5701). Studies also found that abacavir-induced ADRs were seldom observed in patients carrying the HLA-B*5801 subtype. HLA-B*5801 of the same serotype (B17) as B*5701 differs by only 4 amino acids from B*5701. It is believed that because of these sequence differences, HLA-B*5801 cannot bind the specific peptides which are required for HLA-B*5701 to stimulate the T cell immune response. Thus, the difference in peptide binding profiles between HLA-B*5701 and B*5801 is an important clue for exploring the mechanisms of abacavir-induced ADRs. VHSE (principal component score vector of hydrophobic, steric, and electronic properties), a set of amino acid structural descriptors, was employed to establish QSAR models of peptide-binding affinities of HLA-B*5701 and B*5801. Optimal linear SVM (support vector machine) models with high predictive capabilities were obtained for both B*5701 and B*5801. The R2 (coefficient of determination), Q2 (cross-validated R2), and RPRE2 (R2 of test set) of two optimal models were 0.7530, 0.7037, 0.6153 (B*5701) and 0.6074, 0.5966, 0.5762 (B*5801), respectively. For B*5701 and B*5801, the mutations in positions 45 (MET-THR) and 46 (ALA-GLU) have little influence on the selection specificity of the P2 position of the bound peptide. However, the mutation in position 97 (VAL-ARG) greatly influences the selection specificity of the P7 position. HLA-B*5701 prefers the bulky and positively charged amino acids at the P7 position. In contrast, HLA-B*5801 prefers the non-polar hydrophobic amino acids at the P7 position while positively charged amino acids are unfavored.

Surflex-Dock is employed to investigate interactions between neuraminidase inhibitors (NIs) and neuraminidase (NA), which illuminate that carboxyl group, amino (guanidino) group, amide group, hydroxy group are crucial. Hydrogen bonds and hydrophobic interactions impact on activities of NIs. There is a strong correlation between binding affinity and pIC50, with r = 0.813. We have developed three-dimensional holographic vector of atomic interaction field analysis (HoVAIFA) as a new method of 3D-QSAR to understand chemical–biological interactions. Good results, R2 = 0.789 and R2cv = 0.732, show that HoVAIFA can be applicable to molecular structural characterization and bioactivity prediction. Electrostatic, steric and hydrophobic interactions affect activities of NIs. HoVAIFA and docking results are corresponding, which illustrates that HoVAIFA is an effective methodology for characterization of complex interactions of drug molecules.Correlation plot of experimental and calculated pEC50 elucidates that almost all samples are uniformly distributed in a straight line around 45° origin. 3D-HoVAIF can illustrate structural feature of compounds.

Surflex-Dock was applied to study interactions between 30 thiourea analogs and neuraminidase (NA). The docking results showed that hydrogen bonding and electrostatic interactions were highly correlated with the activities of neuraminidase inhibitors (NIs), followed by hydrophobic and steric factors. Moreover, there was a strong correlation between the predicted binding affinity (total score) and experimental pIC50 (correlation coefficient r = 0.870; P < 0.0001). A three dimensional holographic vector of atomic interaction field (3D-HoVAIF) was employed to construct a QSAR model. The r2, q2 and r2test values of the optimal QSAR model were 0.849, 0.724 and 0.689, respectively. From the QSAR model, it could be seen that electrostatic, hydrophobic and steric interactions were closely related to inhibitory activity, which was consistent with the docking results. Based on the docking and QSAR results, five new compounds with high predicted activities were designed.

•More than 12,000 compounds are used to establish prediction models.•A simple interpretable SVM model with strong predictive power is established.•The overall predictive accuracy is larger than 0.80.•Combined fragment descriptors are extremely useful in screening CYP inhibitors.The human cytochrome P450 (CYP450) superfamily plays an important role in drug–drug interactions, drug metabolism, and toxicity. Therefore, prediction of CYP450 inhibitors is extremely important in drug discovery and personal medicine. In this paper, characterized by fragment-based molecular hologram and MACCS descriptors, over 12,000 unique compounds with known CYP2C19 inhibitory activities were used to develop prediction models by partial least squares discriminant analysis (PLSDA) and support vector machine (SVM) methods. By combining two types of fragment-based descriptors, an optimal SVM model with an RBF kernel was obtained. The sensitivity, specificity, accuracy, and Matthews correlation coefficient (MCC) were 86.54%, 83.39%, 84.76%, and 0.6946 for a training set (n = 5,387), and 83.19%, 78.82%, 80.72%, and 0.6152 for a test set (n = 5,383), respectively. The optimal SVM model was further validated by an independent dataset (n = 1,470) with an overall accuracy of 82.38%. The results showed that these two types of fragment-based descriptors are, in some degree, complementary to each other, and can be combined to enhance model predictive power. In comparison with other 2-D or 3-D description methods, the combination of fragment descriptors seems extremely useful in constructing high-throughput screening models of CYP inhibitors in the process of drug discovery.

P-glycoprotein (P-gp), an ATP-binding cassette (ABC) multidrug transporter, can actively transport a broad spectrum of chemically diverse substrates out of cells and is heavily involved in multidrug resistance (MDR) in tumors. So far, the multiple specific binding sites remain a major obstacle in developing an efficient prediction method for P-gp substrates. Herein, emerging chemical pattern (ECP) combined by hierarchical cluster analysis was utilized to predict P-gp substrates as well as their potential binding sites. An optimal ECP model using only 3 descriptors was established with prediction accuracies of 0.80, 0.81 and 0.74 for 803 training samples, 120 test samples, and 179 independent validation samples, respectively. Hierarchical cluster analysis (HCA) of the ECPs of P-gp substrates derived 2 distinct ECP groups (ECPGs). Interestingly, HCA of the P-gp substrates based on ECP similarities also showed 2 distinct classes, which happened to be dominated by the 2 ECPGs, respectively. In the light of available experimental proofs and molecular docking results, the 2 distinct ECPGs were proved to be closely related to the binding profiles of R- and H-site substrates, respectively. The present study demonstrates, for the first time, a successful ECP model, which can not only accurately predict P-gp substrates, but also identify their potential substrate-binding sites.Download high-res image (168KB)Download full-size image