Protein Descriptors Research Articles

Illuminating the origins of spectral properties of green fluorescent proteins via proteochemometric and molecular modeling.

Green fluorescent protein (GFP) has immense utility in biomedical imaging owing to its autofluorescent nature. In efforts to broaden the spectral diversity of GFP, there have been several reports of engineered mutants via rational design and random mutagenesis. Understanding the origins of spectral properties of GFP could be achieved by means of investigating its structure-activity relationship. The first quantitative structure-property relationship study for modeling the spectral properties, particularly the excitation and emission maximas, of GFP was previously proposed by us some years ago in which quantum chemical descriptors were used for model development. However, such simplified model does not consider possible effects that neighboring amino acids have on the conjugated π-system of GFP chromophore. This study describes the development of a unified proteochemometric model in which the GFP chromophore and amino acids in its vicinity are both considered in the same model. The predictive performance of the model was verified by internal and external validation as well as Y-scrambling. Our strategy provides a general solution for elucidating the contribution that specific ligand and protein descriptors have on the investigated spectral property, which may be useful in engineering novel GFP variants with desired characteristics.

Prediction of protein structure classes by incorporating different protein descriptors into general Chou’s pseudo amino acid composition

Successful protein structure identification enables researchers to estimate the biological functions of proteins, yet it remains a challenging problem. The most common method for determining an unknown protein’s structural class is to perform expensive and time-consuming manual experiments. Because of the availability of amino acid sequences generated in the post-genomic age, it is possible to predict an unknown protein’s structural class using machine learning methods given a protein’s amino-acid sequence and/or its secondary structural elements. Following recent research in this area, we propose a new machine learning system that is based on combining several protein descriptors extracted from different protein representations, such as position specific scoring matrix (PSSM), the amino-acid sequence, and secondary structural sequences. The prediction engine of our system is operated by an ensemble of support vector machines (SVMs), where each SVM is trained on a different descriptor. The results of each SVM are combined by sum rule. Our final ensemble produces a success rate that is substantially better than previously reported results on three well-established datasets. The MATLAB code and datasets used in our experiments are freely available for future comparison at http://www.dei.unipd.it/node/2357.

A set of descriptors for identifying the protein–drug interaction in cellular networking

The study of protein–drug interactions is a significant issue for drug development. Unfortunately, it is both expensive and time-consuming to perform physical experiments to determine whether a drug and a protein are interacting with each other. Some previous attempts to design an automated system to perform this task were based on the knowledge of the 3D structure of a protein, which is not always available in practice. With the availability of protein sequences generated in the post-genomic age, however, a sequence-based solution to deal with this problem is necessary. Following other works in this area, we propose a new machine learning system based on several protein descriptors extracted from several protein representations, such as, variants of the position specific scoring matrix (PSSM) of proteins, the amino-acid sequence, and a matrix representation of a protein. The prediction engine is operated by an ensemble of support vector machines (SVMs), with each SVM trained on a specific descriptor and the results of each SVM combined by sum rule. The overall success rate achieved by our final ensemble is notably higher than previous results obtained on the same datasets using the same testing protocols reported in the literature.MATLAB code and the datasets used in our experiments are freely available for future comparison at http://www.dei.unipd.it/node/2357.

Classification of multi-family enzymes by multi-label machine learning and sequence-based descriptors

Multi-family enzymes are of great importance in life, disease and other domains. However, in terms of the classification of enzymes, the information of multi-family enzymes is always removed from the dataset to account for the limitation of traditional single-label prediction methods. In order to predict multiple classes of multi-family enzymes, we adopted two multi-label learning algorithms, namely RAkEL-RF and MLKNN, and two types of protein descriptors, namely CTD and PseAAC, to generate four predictors, RAkEL-RF-CTD, RAkEL-RF-PseAAC, MLKNN-CTD and MLKNN-PseAAC. When the four predictors were tested on a training set with 10-fold cross validation, the overall success rates reached 97.99%, 96.07%, 96.01% and 95.31%, respectively. For the independent test set, the corresponding rates reached 97.57%, 95.03%, 95.9% and 93.9%, respectively. In conclusion, it proved the outstanding prediction capability and robustness of our predictors from the extremely small difference between two sets for each predictor and the relatively higher accuracy. In addition, three of seven pairs of homologous enzymes with different functions and eighteen of twenty-three distantly related enzymes with a similar family were correctly classified by the RAkEL-RF-CTD predictor. These results indicated the extensive applicability of our predictors.

Computational chemogenomics: is it more than inductive transfer?

High-throughput assays challenge us to extract knowledge from multi-ligand, multi-target activity data. In QSAR, weights are statically fitted to each ligand descriptor with respect to a single endpoint or target. However, computational chemogenomics (CG) has demonstrated benefits of learning from entire grids of data at once, rather than building target-specific QSARs. A possible reason for this is the emergence of inductive knowledge transfer (IT) between targets, providing statistical robustness to the model, with no assumption about the structure of the targets. Relevant protein descriptors in CG should allow one to learn how to dynamically adjust ligand attribute weights with respect to protein structure. Hence, models built through explicit learning (EL) by including protein information, while benefitting from IT enhancement, should provide additional predictive capability, notably for protein deorphanization. This interplay between IT and EL in CG modeling is not sufficiently studied. While IT is likely to occur irrespective of the injected target information, it is not clear whether and when boosting due to EL may occur. EL is only possible if protein description is appropriate to the target set under investigation. The key issue here is the search for evidence of genuine EL exceeding expectations based on pure IT. We explore the problem in the context of Support Vector Regression, using more than 9,400 pKi values of 31 GPCRs, where compound-protein interactions are represented by the concatenation of vectorial descriptions of compounds and proteins. This provides a unified framework to generate both IT-enhanced and potentially EL-enabled models, where the difference is toggled by supplied protein information. For EL-enabled models, protein information includes genuine protein descriptors such as typical sequence-based terms, but also the experimentally determined affinity cross-correlation fingerprints. These latter benchmark the expected behavior of a quasi-ideal descriptor capturing the actual functional protein-protein relatedness, and therefore thought to be the most likely to enable EL. EL- and IT-based methods were benchmarked alongside classical QSAR, with respect to cross-validation and deorphanization challenges. A rational method for projecting benchmarked methodologies into a strategy space is given, in the aims that the projection will provide directions for the types of molecule designs possible using a given methodology. While EL-enabled strategies outperform classical QSARs and favorably compare to similar published results, they are, in all respects evaluated herein, not strongly distinguished from IT-enhanced models. Moreover, EL-enabled strategies failed to prove superior in deorphanization challenges. Therefore, this paper raises caution that, contrary to common belief and intuitive expectation, the benefits of chemogenomics models over classical QSAR are quite possibly due less to the injection of protein-related information, and rather impacted more by the effect of inductive transfer, due to simultaneous learning from all of the modeled endpoints. These results show that the field of protein descriptor research needs further improvements to truly realize the expected benefit of EL.

Application of orthogonal L-shaped PLS to chemogenomic data and its chemical interpretation from predictive and orthogonal latent variables

We carried out a validation study of the orthogonal L-shaped PLS (OLPLS) method using chemogenomic data based on adenosine receptor inhibitor activity measurements. Using OLPLS, the ligand and protein descriptors could be connected to eight adenosine receptor inhibitor activities. The fingerprints representing specific chemical substructures on the ligands were used as the ligand descriptors, while z-scales were used as the protein descriptors. Three clusters were observed in the chemical and protein spaces from the predictive scores and loadings. From these, the predictive and orthogonal ligand structure fragments towards three adenosine receptors could be successfully elucidated. The predictive fragment for the human adenosine 2A receptor was confirmed by comparison to the X-ray crystal structure. As expected, the orthogonal fragments contained no physicochemical features required for specific interaction with the adenosine receptors.

Markov mean properties for cell death-related protein classification

The cell death (CD) is a dynamic biological function involved in physiological and pathological processes. Due to the complexity of CD, there is a demand for fast theoretical methods that can help to find new CD molecular targets. The current work presents the first classification model to predict CD-related proteins based on Markov Mean Properties. These protein descriptors have been calculated with the MInD-Prot tool using the topological information of the amino acid contact networks of the 2423 protein chains, five atom physicochemical properties and the protein 3D regions. The Machine Learning algorithms from Weka were used to find the best classification model for CD-related protein chains using all 20 attributes. The most accurate algorithm to solve this problem was K*. After several feature subset methods, the best model found is based on only 11 variables and is characterized by the Area Under the Receiver Operating Characteristic Curve (AUROC) of 0.992 and the true positive rate (TP Rate) of 88.2% (validation set). 7409 protein chains labeled with “unknown function” in the PDB Databank were analyzed with the best model in order to predict the CD-related biological activity. Thus, several proteins have been predicted to have CD-related function in Homo sapiens: 3DRX–involved in virus-host interaction biological process, protein homooligomerization; 4DWF–involved in cell differentiation, chromatin modification, DNA damage response, protein stabilization; 1IUR–involved in ATP binding, chaperone binding; 1J7D–involved in DNA double-strand break processing, histone ubiquitination, nucleotide-binding oligomerization; 1UTU–linked with DNA repair, regulation of transcription; 3EEC–participating to the cellular membrane organization, egress of virus within host cell, class mediator resulting in cell cycle arrest, negative regulation of ubiquitin-protein ligase activity involved in mitotic cell cycle and apoptotic process. Other proteins from bacteria predicted as CD-related are 2G3V - a CAG pathogenicity island protein 13 from Helicobacter pylori, 4G5A - a hypothetical protein in Bacteroides thetaiotaomicron, 1YLK–involved in the nitrogen metabolism of Mycobacterium tuberculosis, and 1XSV - with possible DNA/RNA binding domains. The results demonstrated the possibility to predict CD-related proteins using molecular information encoded into the protein 3D structure. Thus, the current work demonstrated the possibility to predict new molecular targets involved in cell-death processes.

Improving enzyme regulatory protein classification by means of SVM-RFE feature selection

Enzyme regulation proteins are very important due to their involvement in many biological processes that sustain life. The complexity of these proteins, the impossibility of identifying direct quantification molecular properties associated with the regulation of enzymatic activities, and their structural diversity creates the necessity for new theoretical methods that can predict the enzyme regulatory function of new proteins. The current work presents the first classification model that predicts protein enzyme regulators using the Markov mean properties. These protein descriptors encode the topological information of the amino acid into contact networks based on amino acid distances and physicochemical properties. MInD-Prot software calculated these molecular descriptors for 2415 protein chains (350 enzyme regulators) using five atom physicochemical properties (Mulliken electronegativity, Kang-Jhon polarizability, vdW area, atom contribution to P) and the protein 3D regions. The best classification models to predict enzyme regulators have been obtained with machine learning algorithms from Weka using 18 features. K* has been demonstrated to be the most accurate algorithm for this protein function classification. Wrapper Subset Evaluator and SVM-RFE approaches were used to perform a feature subset selection with the best results obtained from SVM-RFE. Classification performance employing all the available features can be reached using only the 8 most relevant features selected by SVM-RFE. Thus, the current work has demonstrated the possibility of predicting new molecular targets involved in enzyme regulation using fast theoretical algorithms.

L-shaped PLS analysis of multiple inhibitory activities of adrenergic alpha receptors using ligand and protein matrices

Recently, chemogenomics has been given high attention in pharmaceutical industry. By definition, chemogenomics data is a two-dimensional matrix, where proteins are usually reported as columns and ligands as rows, and where reported values are usually inhibitory activities. In a straightforward manner, it would be the best choice that the chemogenomics matrix is explained by two matrices that each consists of ligand and protein descriptors. Bi-modal PLS follows this concept and several variants have been proposed. Among them, we focus on the L-shaped PLS (LPLS) method and apply it to aminergic G protein-coupled receptor inhibitory activity data in a previous study.In this study, we have devised both of the ligand and protein matrices in the frame of LPLS analysis for four adrenergic alpha receptors. In the ligand matrix, the similarity matrix derived from the Tanimoto similarity value between the pair of inhibitors was employed. As for the protein matrix, the 3D protein pocket was mapped to the spherical self-organizing map sphere. Then, the lipophilic potential value on each node was used as protein descriptors. Thanks to four plots of LPLS, we could easily identify four selective inhibitors and elucidate structural requirements for selectivity.

The use of quantitative structure–activity relationship models to develop optimized processes for the removal of tobacco host cell proteins during biopharmaceutical production

The production of recombinant pharmaceutical proteins in plants benefits from the low cost of upstream production and the greater scalability of plants compared to fermenter-based systems. Now that manufacturing processes that comply with current good manufacturing practices have been developed, plants can compete with established platforms on equal terms. However, the costs of downstream processing remain high, in part because of the dedicated process steps required to remove plant-specific process-related impurities. We therefore investigated whether the ideal strategy for the chromatographic removal of tobacco host cell proteins can be predicted by quantitative structure–activity relationship (QSAR) modeling to reduce the process development time and overall costs. We identified more than 100 tobacco proteins by mass spectrometry and their structures were reconstructed from X-ray crystallography, nuclear magnetic resonance spectroscopy and/or homology modeling data. The resulting three-dimensional models were used to calculate protein descriptors, and significant descriptors were selected based on recently-published retention data for model proteins to develop QSAR models for protein retention on anion, cation and mixed-mode resins. The predicted protein retention profiles were compared with experimental results using crude tobacco protein extracts. Because of the generic nature of the method, it can easily be transferred to other expression systems such as mammalian cells. The quality of the models and potential improvements are discussed.

Benchmarking of protein descriptor sets in proteochemometric modeling (part 2): modeling performance of 13 amino acid descriptor sets

BackgroundWhile a large body of work exists on comparing and benchmarking descriptors of molecular structures, a similar comparison of protein descriptor sets is lacking. Hence, in the current work a total of 13 amino acid descriptor sets have been benchmarked with respect to their ability of establishing bioactivity models. The descriptor sets included in the study are Z-scales (3 variants), VHSE, T-scales, ST-scales, MS-WHIM, FASGAI, BLOSUM, a novel protein descriptor set (termed ProtFP (4 variants)), and in addition we created and benchmarked three pairs of descriptor combinations. Prediction performance was evaluated in seven structure-activity benchmarks which comprise Angiotensin Converting Enzyme (ACE) dipeptidic inhibitor data, and three proteochemometric data sets, namely (1) GPCR ligands modeled against a GPCR panel, (2) enzyme inhibitors (NNRTIs) with associated bioactivities against a set of HIV enzyme mutants, and (3) enzyme inhibitors (PIs) with associated bioactivities on a large set of HIV enzyme mutants.ResultsThe amino acid descriptor sets compared here show similar performance (<0.1 log units RMSE difference and <0.1 difference in MCC), while errors for individual proteins were in some cases found to be larger than those resulting from descriptor set differences ( > 0.3 log units RMSE difference and >0.7 difference in MCC). Combining different descriptor sets generally leads to better modeling performance than utilizing individual sets. The best performers were Z-scales (3) combined with ProtFP (Feature), or Z-Scales (3) combined with an average Z-Scale value for each target, while ProtFP (PCA8), ST-Scales, and ProtFP (Feature) rank last.ConclusionsWhile amino acid descriptor sets capture different aspects of amino acids their ability to be used for bioactivity modeling is still – on average – surprisingly similar. Still, combining sets describing complementary information consistently leads to small but consistent improvement in modeling performance (average MCC 0.01 better, average RMSE 0.01 log units lower). Finally, performance differences exist between the targets compared thereby underlining that choosing an appropriate descriptor set is of fundamental for bioactivity modeling, both from the ligand- as well as the protein side.

Benchmarking of protein descriptor sets in proteochemometric modeling (part 1): comparative study of 13 amino acid descriptor sets

BackgroundWhile a large body of work exists on comparing and benchmarking of descriptors of molecular structures, a similar comparison of protein descriptor sets is lacking. Hence, in the current work a total of 13 different protein descriptor sets have been compared with respect to their behavior in perceiving similarities between amino acids. The descriptor sets included in the study are Z-scales (3 variants), VHSE, T-scales, ST-scales, MS-WHIM, FASGAI and BLOSUM, and a novel protein descriptor set termed ProtFP (4 variants). We investigate to which extent descriptor sets show collinear as well as orthogonal behavior via principal component analysis (PCA).ResultsIn describing amino acid similarities, MSWHIM, T-scales and ST-scales show related behavior, as do the VHSE, FASGAI, and ProtFP (PCA3) descriptor sets. Conversely, the ProtFP (PCA5), ProtFP (PCA8), Z-Scales (Binned), and BLOSUM descriptor sets show behavior that is distinct from one another as well as both of the clusters above. Generally, the use of more principal components (>3 per amino acid, per descriptor) leads to a significant differences in the way amino acids are described, despite that the later principal components capture less variation per component of the original input data.ConclusionIn this work a comparison is provided of how similar (and differently) currently available amino acids descriptor sets behave when converting structure to property space. The results obtained enable molecular modelers to select suitable amino acid descriptor sets for structure-activity analyses, e.g. those showing complementary behavior.

Large-scale prediction of human kinase–inhibitor interactions using protein sequences and molecular topological structures

The kinase family is one of the largest target families in the human genome. The family's key function in signal transduction for all organisms makes it a very attractive target class for the therapeutic interventions in many diseases states such as cancer, diabetes, inflammation and arthritis. A first step toward accelerating kinase drug discovery process is to fast identify whether a chemical and a kinase interact or not. Experimentally, these interactions can be identified by in vitro binding assay – an expensive and laborious procedure that is not applicable on a large scale. Therefore, there is an urgent need to develop statistically efficient approaches for identifying kinase–inhibitor interactions. For the first time, the quantitative binding affinities of kinase–inhibitor pairs are differentiated as a measurement to define if an inhibitor interacts with a kinase, and then a chemogenomics framework using an unbiased set of general integrated features (drug descriptors and protein descriptors) and random forest (RF) is employed to construct a predictive model which can accurately classify kinase–inhibitor pairs. Our results show that RF with integrated features gave prediction accuracy of 93.76%, sensitivity of 92.26%, and specificity of 95.27%, respectively. The results are superior to those by only considering two separated spaces (chemical space and protein space), demonstrating that these integrated features contribute cooperatively. Based on the constructed model, we provided a high confidence list of drug–target associations for subsequent experimental investigation guidance at a low false discovery rate.

Study on human GPCR–inhibitor interactions by proteochemometric modeling

G protein-coupled receptors (GPCRs) are the most frequently addressed drug targets in the pharmaceutical industry. However, achieving highly safety and efficacy in designing of GPCR drugs is quite challenging since their primary amino acid sequences show fairly high homology. Systematic study on the interaction spectra of inhibitors with multiple human GPCRs will shed light on how to design the inhibitors for different diseases which are related to GPCRs. To reach this goal, several proteochemometric models were constructed based on different combinations of two protein descriptors, two ligand descriptors and one ligand–receptor cross-term by two kinds of statistical learning techniques.Our results show that support vector regression (SVR) performs better than Gaussian processes (GP) for most combinations of descriptors. The transmembrane (TM) identity descriptors have more powerful ability than the z-scale descriptors in the characterization of GPCRs. Furthermore, the performance of our PCM models was not improved by introducing the cross-terms. Finally, based on the TM Identity descriptors and 28-dimensional drug-like index, two best PCM models with GP and SVR (GP-S-DLI: R2=0.9345, Q2test=0.7441; SVR-S-DLI: R2=1.0000, Q2test=0.7423) were derived respectively. The area of ROC curve was 0.8940 when an external test set was used, which indicates that our PCM model obtained a powerful capability for predicting new interactions between GPCRs and ligands.Our results indicate that the derived best model has a high predictive ability for human GPCR–inhibitor interactions. It can be potentially used to discover novel multi-target or specific inhibitors of GPCRs with higher efficacy and fewer side effects.

Prediction of the functional consequences of single amino acid substitution in human cytochrome P450

Non-synonymous single nucleotide polymorphisms (nsSNPs) which cause single amino acid substitution in cytochrome P450 have been regarded as a considerable factor in determining the capability of drug metabolism. The examination of functional consequences of nsSNPs is of great importance in evaluating different drug metabolism capabilities among individuals and developing the personalised medicine. In this study, we use a support vector machine combined with genetic algorithm to construct the predictive model, whereby we can predict the functional consequences of nsSNPs of nine major human cytochrome P450s. In addition, we develop a strategy to calculate two kinds of protein descriptors using the amino acid indices. In contrast to the amino acid indices, the protein descriptors can fully consider the change caused by the substitution of single amino acid. After multi-round feature selection, five protein descriptors and eight substrate descriptors are selected to construct the final model. The validation results exhibit that our model can achieve high predictive accuracy (Acc) of 88.1%, sensitivity of 93.1% and specificity of 72.7%. Overall, the main improvement of our model over other existing models is the relatively high Acc and the consideration of the properties of protein and substrate in the prediction.

Computational biology of RNA interactions

The biodiversity of the RNA world has been underestimated for decades. RNA molecules are key building blocks, sensors, and regulators of modern cells. The biological function of RNA molecules cannot be separated from their ability to bind to and interact with a wide space of chemical species, including small molecules, nucleic acids, and proteins. Computational chemists, physicists, and biologists have developed a rich tool set for modeling and predicting RNA interactions. These interactions are to some extent determined by the binding conformation of the RNA molecule. RNA binding conformations are approximated with often acceptable accuracy by sequence and secondary structure motifs. Secondary structure ensembles of a given RNA molecule can be efficiently computed in many relevant situations by employing a standard energy model for base pair interactions and dynamic programming techniques. The case of bi-molecular RNA-RNA interactions can be seen as an extension of this approach. However, unbiased transcriptome-wide scans for local RNA-RNA interactions are computationally challenging yet become efficient if the binding motif/mode is known and other external information can be used to confine the search space. Computational methods are less developed for proteins and small molecules, which bind to RNA with very high specificity. Binding descriptors of proteins are usually determined by in vitro high-throughput assays (e.g., microarrays or sequencing). Intriguingly, recent experimental advances, which are mostly based on light-induced cross-linking of binding partners, render in vivo binding patterns accessible yet require new computational methods for careful data interpretation. The grand challenge is to model the in vivo situation where a complex interplay of RNA binders competes for the same target RNA molecule. Evidently, bioinformaticians are just catching up with the impressive pace of these developments.

An empirical study on the matrix-based protein representations and their combination with sequence-based approaches

Many domains have a stake in the development of reliable systems for automatic protein classification. Of particular interest in recent studies of automatic protein classification is the exploration of new methods for extracting features from a protein that enhance classification for specific problems. These methods have proven very useful in one or two domains, but they have failed to generalize well across several domains (i.e. classification problems). In this paper, we evaluate several feature extraction approaches for representing proteins with the aim of sequence-based protein classification. Several protein representations are evaluated, those starting from: the position specific scoring matrix (PSSM) of the proteins; the amino-acid sequence; a matrix representation of the protein, of dimension (length of the protein) ×20, obtained using the substitution matrices for representing each amino-acid as a vector. A valuable result is that a texture descriptor can be extracted from the PSSM protein representation which improves the performance of standard descriptors based on the PSSM representation. Experimentally, we develop our systems by comparing several protein descriptors on nine different datasets. Each descriptor is used to train a support vector machine (SVM) or an ensemble of SVM. Although different stand-alone descriptors work well on some datasets (but not on others), we have discovered that fusion among classifiers trained using different descriptors obtains a good performance across all the tested datasets. Matlab code/Datasets used in the proposed paper are available at http://www.bias.csr.unibo.it\nanni\PSSM.rar.

Survey of the camel urinary proteome by shotgun proteomics using a multiple database search strategy

We report the first survey of the dromedary camel urinary proteome. Proteins retained from ultrafiltration of urine were analysed by GeLC-MS/MS (SDS-PAGE followed by LC-MS/MS). In the absence of a complete camel genome sequence, the number of protein identifications was maximised by searching three primary sequence databases: Swiss-Prot, alpaca and camel EST. This search strategy enabled the identification of 1274 peptide sequences, of which 735 were found in at least two independent samples. Functional annotations for proteins identified from alpaca and camel EST sequences were mapped from basic local alignment search tool (protein) searches. These 735 peptides, which included many novel sequences found only in the camel EST database, were grouped to 147 protein descriptors. Gene ontology term analysis of human proteins with sequence similarity showed that camel urine may be particularly enriched in proteins from extracellular compartments and vesicles, and with functions that include carbohydrate-binding and peptidase inhibitor activities. If their biological functions are conserved between species, many of the camel urinary proteins could be involved in various stress and immune responses, and some may have antimicrobial activities.

Screening of selective histone deacetylase inhibitors by proteochemometric modeling

BackgroundHistone deacetylase (HDAC) is a novel target for the treatment of cancer and it can be classified into three classes, i.e., classes I, II, and IV. The inhibitors selectively targeting individual HDAC have been proved to be the better candidate antitumor drugs. To screen selective HDAC inhibitors, several proteochemometric (PCM) models based on different combinations of three kinds of protein descriptors, two kinds of ligand descriptors and multiplication cross-terms were constructed in our study.ResultsThe results show that structure similarity descriptors are better than sequence similarity descriptors and geometry descriptors in the leftacterization of HDACs. Furthermore, the predictive ability was not improved by introducing the cross-terms in our models. Finally, a best PCM model based on protein structure similarity descriptors and 32-dimensional general descriptors was derived (R2 = 0.9897, Qtest2 = 0.7542), which shows a powerful ability to screen selective HDAC inhibitors.ConclusionsOur best model not only predict the activities of inhibitors for each HDAC isoform, but also screen and distinguish class-selective inhibitors and even more isoform-selective inhibitors, thus it provides a potential way to discover or design novel candidate antitumor drugs with reduced side effect.

Proteochemometric Modeling of the Bioactivity Spectra of HIV-1 Protease Inhibitors by Introducing Protein-Ligand Interaction Fingerprint

HIV-1 protease is one of the main therapeutic targets in HIV. However, a major problem in treatment of HIV is the rapid emergence of drug-resistant strains. It should be particularly helpful to clinical therapy of AIDS if one method can be used to predict antivirus capability of compounds for different variants. In our study, proteochemometric (PCM) models were created to study the bioactivity spectra of 92 chemical compounds with 47 unique HIV-1 protease variants. In contrast to other PCM models, which used Multiplication of Ligands and Proteins Descriptors (MLPD) as cross-term, one new cross-term, i.e. Protein-Ligand Interaction Fingerprint (PLIF) was introduced in our modeling. With different combinations of ligand descriptors, protein descriptors and cross-terms, nine PCM models were obtained, and six of them achieved good predictive abilities (Q2 test>0.7). These results showed that the performance of PCM models could be improved when ligand and protein descriptors were complemented by the newly introduced cross-term PLIF. Compared with the conventional cross-term MLPD, the newly introduced PLIF had a better predictive ability. Furthermore, our best model (GD & P & PLIF: Q2test = 0.8271) could select out those inhibitors which have a broad antiviral activity. As a conclusion, our study indicates that proteochemometric modeling with PLIF as cross-term is a potential useful way to solve the HIV-1 drug-resistant problem.