Drug Activity Prediction Research Articles

Implementation of multiple-instance learning in drug activity prediction

BackgroundIn the context of drug discovery and development, much effort has been exerted to determine which conformers of a given molecule are responsible for the observed biological activity. In this work we aimed to predict bioactive conformers using a variant of supervised learning, named multiple-instance learning. A single molecule, treated as a bag of conformers, is biologically active if and only if at least one of its conformers, treated as an instance, is responsible for the observed bioactivity; and a molecule is inactive if none of its conformers is responsible for the observed bioactivity. The implementation requires instance-based embedding, and joint feature selection and classification. The goal of the present project is to implement multiple-instance learning in drug activity prediction, and subsequently to identify the bioactive conformers for each molecule.MethodsWe encoded the 3-dimensional structures using pharmacophore fingerprints which are binary strings, and accomplished instance-based embedding using calculated dissimilarity distances. Four dissimilarity measures were employed and their performances were compared. 1-norm SVM was used for joint feature selection and classification. The approach was applied to four data sets, and the best proposed model for each data set was determined by using the dissimilarity measure yielding the smallest number of selected features.ResultsThe predictive abilities of the proposed approach were compared with three classical predictive models without instance-based embedding. The proposed approach produced the best predictive models for one data set and second best predictive models for the rest of the data sets, based on the external validations. To validate the ability of the proposed approach to find bioactive conformers, 12 small molecules with co-crystallized structures were seeded in one data set. 10 out of 12 co-crystallized structures were indeed identified as significant conformers using the proposed approach.ConclusionsThe proposed approach was proven not to suffer from overfitting and to be highly competitive with classical predictive models, so it is very powerful for drug activity prediction. The approach was also validated as a useful method for pursuit of bioactive conformers.

Combined rule extraction and feature elimination in supervised classification.

There are a vast number of biology related research problems involving a combination of multiple sources of data to achieve a better understanding of the underlying problems. It is important to select and interpret the most important information from these sources. Thus it will be beneficial to have a good algorithm to simultaneously extract rules and select features for better interpretation of the predictive model. We propose an efficient algorithm, Combined Rule Extraction and Feature Elimination (CRF), based on 1-norm regularized random forests. CRF simultaneously extracts a small number of rules generated by random forests and selects important features. We applied CRF to several drug activity prediction and microarray data sets. CRF is capable of producing performance comparable with state-of-the-art prediction algorithms using a small number of decision rules. Some of the decision rules are biologically significant.

Feature selection in multi-instance learning

Multi-instance learning was first proposed by Dietterich et al. (Artificial Intelligence 89(1–2):31–71, 1997) when they were investigating the problem of drug activity prediction. Here, the training set is composed of labeled bags, each of which consists of many unlabeled instances. And the goal of this learning framework is to learn some classifier from the training set for correctly labeling unseen bags. After Dietterich et al., many studies about this new learning framework have been started and many new algorithms have been proposed, for example, DD, EM-DD, Citation-kNN and so on. All of these algorithms are working on the full data set. But as in single-instance learning, different feature in training set has different effect on the training about classifier. In this paper, we will study the problem about feature selection in multi-instance learning. We will extend the data reliability measure and make it select the key feature in multi-instance scenario.

3-D QSAutogrid/R: An Alternative Procedure To Build 3-D QSAR Models. Methodologies and Applications

Since it first appeared in 1988 3-D QSAR has proved its potential in the field of drug design and activity prediction. Although thousands of citations now exist in 3-D QSAR, its development was rather slow with the majority of new 3-D QSAR applications just extensions of CoMFA. An alternative way to build 3-D QSAR models, based on an evolution of software, has been named 3-D QSAutogrid/R and has been developed to use only software freely available to academics. 3-D QSAutogrid/R covers all the main features of CoMFA and GRID/GOLPE with implementation by multiprobe/multiregion variable selection (MPGRS) that improves the simplification of interpretation of the 3-D QSAR map. The methodology is based on the integration of the molecular interaction fields as calculated by AutoGrid and the R statistical environment that can be easily coupled with many free graphical molecular interfaces such as UCSF-Chimera, AutoDock Tools, JMol, and others. The description of each R package is reported in detail, and, to assess its validity, 3-D QSAutogrid/R has been applied to three molecular data sets of which either CoMFA or GRID/GOLPE models were reported in order to compare the results. 3-D QSAutogrid/R has been used as the core engine to prepare more that 240 3-D QSAR models forming the very first 3-D QSAR server ( www.3d-qsar.com ) with its code freely available through R-Cran distribution.

Large-scale prediction and testing of drug activity on side-effect targets

SummaryDiscovering the unintended “off-targets” that predict adverse drug reactions (ADRs) is daunting by empirical methods alone. Drugs can act on multiple protein targets, some of which can be unrelated by traditional molecular metrics, and hundreds of proteins have been implicated in side effects. We therefore explored a computational strategy to predict the activity of 656 marketed drugs on 73 unintended “side effect” targets. Approximately half of the predictions were confirmed, either from proprietary databases unknown to the method or by new experimental assays. Affinities for these new off-targets ranged from 1 nM to 30 μM. To explore relevance, we developed an association metric to prioritize those new off-targets that explained side effects better than any known target of a given drug, creating a Drug-Target-ADR network. Among these new associations was the prediction that the abdominal pain side effect of the synthetic estrogen chlorotrianisene was mediated through its newly discovered inhibition of the enzyme COX-1. The clinical relevance of this inhibition was borne-out in whole human blood platelet aggregation assays. This approach may have wide application to de-risking toxicological liabilities in drug discovery.

Computational strategy for predicting drug activity on side-effect targets

Computational strategy for predicting drug activity on side-effect targets

Abstract LB-455: Reciprocal expression of miR125b and miR192/miR200 families define three clinically relevant sub-types of colorectal cancer with differential sensitivity to EGFR and MEK targeted agents

Abstract Global miRNA profiles are particularly informative regarding cell lineage and the differentiation state of tumours. Recent studies also indicate specific miRNAs are upregulated in following drug treatment and in poor prognosis or metastatic disease. The aims of this study were to determine if miRNA profiles could be used to segment a 50 colorectal cell line panel into clinically relevant sub-types and to determine if the miRNA data enhanced prediction of pharmacological response. The global miRNA profile was examined using unsupervised two dimensional hierarchical clustering which revealed a small group of miRNAs which displayed reciprocal expression between the cell line clusters. A subset of 9 miRNAs, miR192 cluster (3 miRs), miR125b (3miRs) cluster and a miR200 negative cluster (3miRs) defined three cell sub-types. These miRs have been associated with intestinal differentiation, poor prognosis/metastasis in CRC and mesenchymal transition respectively. Importantly, the cell sub-types defined by miRNAs directly relate to sub-types defined by iterative clustering of mRNA data in clinical CRC tissue. Analysis of differentially expressed genes between the cell sub-types revealed the expected differences in EMT genes between the miR200 negative cluster and the miR125b and miR192 clusters. Interestingly, the miR192 positive sub-type was discriminated from the miR125b positive cluster by higher expression of several genes specifically associated with intestinal differentiation including CDX1 (71-fold), HNF1a (TCF1) (3-fold), HNF4a (TCF14) (10-fold), HNF4G (5-fold) ISX (12-fold), and IHH (24-fold). Significantly, there were also differences in mRNA expression of genes related to drug sensitivity e.g. ERBB3 (3.6-fold), AREG (3-fold). In contrast, the miR125b subtype had high expression of IRX3 (12-fold) a homeobox gene associated with neural differentiation and increased expression of the oncogene Axl (6-fold). Protein data revealed the miR192 positive sub-type also had higher expression of ERBB2 and B-catenin and lower levels of phosphorylation of AKT substrate, p70S6K and S6 when compared to the miR200 negative and miR125b clusters. The miR192 sub-type also exhibited greater sensitivity to ERB and MEK targeted agents compared to the miR200 and miR125b sub-types. Taken together, the miR data enables segregation by differentiation status, this data enhances prediction of drug activity compared to mutation data alone and is of significance because both miR expression and differentiation status are modulated by prior chemotherapy regimes. Citation Format: {Authors}. {Abstract title} [abstract]. In: Proceedings of the 103rd Annual Meeting of the American Association for Cancer Research; 2012 Mar 31-Apr 4; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2012;72(8 Suppl):Abstract nr LB-455. doi:1538-7445.AM2012-LB-455

Electroencephalogram-based pharmacodynamic measures: a review.

Pharmacokinetics and pharmacodynamics can provide a useful modeling framework for predicting drug activity and can serve as a basis for dose optimization. Determining a suitable biomarker or surrogate measure of drug effect for pharmacodynamic models can be challenging. The electroencephalograph is a widely-available and non-invasive tool for recording brain-wave activity simultaneously from multiple brain regions. Certain drug classes (such as drugs that act on the central nervous system) also generate a reproducible electroencephalogram (EEG) effect. Characterization of such a drug-induced EEG effect can produce pharmacokinetic/pharmacodynamic models useful for titrating drug levels and expediting development of chemically-similar drug analogs. This paper reviews the relevant concepts involved in pharmacokinetic/pharmacodynamic modeling using EEG-based pharmacodynamic measures. In addition, examples of such models for various drugs are organized by drug activity as well as chemical structure and presented.

A review of multi-instance learning assumptions

AbstractMulti-instance (MI) learning is a variant of inductive machine learning, where each learning example contains a bag of instances instead of a single feature vector. The term commonly refers to the supervised setting, where each bag is associated with a label. This type of representation is a natural fit for a number of real-world learning scenarios, including drug activity prediction and image classification, hence many MI learning algorithms have been proposed. Any MI learning method must relate instances to bag-level class labels, but many types of relationships between instances and class labels are possible. Although all early work in MI learning assumes a specific MI concept class known to be appropriate for a drug activity prediction domain; this ‘standard MI assumption’ is not guaranteed to hold in other domains. Much of the recent work in MI learning has concentrated on a relaxed view of the MI problem, where the standard MI assumption is dropped, and alternative assumptions are considered instead. However, often it is not clearly stated what particular assumption is used and how it relates to other assumptions that have been proposed. In this paper, we aim to clarify the use of alternative MI assumptions by reviewing the work done in this area.

Unified QSAR & network‐based computational chemistry approach to antimicrobials. II. Multiple distance and triadic census analysis of antiparasitic drugs complex networks

In the previous work, we reported a multitarget Quantitative Structure-Activity Relationship (mt-QSAR) model to predict drug activity against different fungal species. This mt-QSAR allowed us to construct a drug-drug multispecies Complex Network (msCN) to investigate drug-drug similarity (González-Díaz and Prado-Prado, J Comput Chem 2008, 29, 656). However, important methodological points remained unclear, such as follows: (1) the accuracy of the methods when applied to other problems; (2) the effect of the distance type used to construct the msCN; (3) how to perform the inverse procedure to study species-species similarity with multidrug resistance CNs (mdrCN); and (4) the implications and necessary steps to perform a substructural Triadic Census Analysis (TCA) of the msCN. To continue the present series with other important problem, we developed here a mt-QSAR model for more than 700 drugs tested in the literature against different parasites (predicting antiparasitic drugs). The data were processed by Linear Discriminate Analysis (LDA) and the model classifies correctly 93.62% (1160 out of 1239 cases) in training. The model validation was carried out by means of external predicting series; the model classified 573 out of 607, that is, 94.4% of cases. Next, we carried out the first comparative study of the topology of six different drug-drug msCNs based on six different distances such as Euclidean, Chebychev, Manhattan, etc. Furthermore, we compared the selected drug-drug msCN and species-species mdsCN with random networks. We also introduced here the inverse methodology to construct species-species msCN based on a mt-QSAR model. Last, we reported the first substructural analysis of drug-drug msCN using Triadic Census Analysis (TCA) algorithm.

Epigenetics and breast cancer – prediction of drug activity

Epigenetics and breast cancer – prediction of drug activity

Immobilized phospholipid capillary electrophoresis for study of drug–membrane interactions and prediction of drug activity

Immobilized phospholipid capillary electrophoresis (IPCE) was developed for studying the interactions between a set of nonsteroidal anti-inflammatory drugs (NSAIDs) and membrane and predicting the biological activity of NSAIDs. Supported vesicle layers and supported phospholipid bilayers were attached to the inner surface of a capillary wall simply by rinsing with liposome solutions. The liposomes, composed of soybean phosphatidylcholine (SPC) or SPC and different proportions of cholesterol (Ch), were small unilamellar vesicles prepared by sonication. The normalized capacity factor ( K IPCE) was introduced into IPCE for evaluating drug–membrane interactions. Related theories and equations were derived to calculate K IPCE values from apparent migration time of a solute and electroosmotic flow. The strong relationships were observed between log K IPCE (SPC) values and log K lw values (the partition coefficients determined in free SPC-liposome partitioning system) ( R = 0.9855 and P < 0.0001) or log K ILC values (the normalized capacity factors determined by immobilized POPC-liposome chromatography, POPC represents 1-palmitoyl-2-oleoyl- sn-glycero-3-phosphocholine) ( R = 0.9875 and P < 0.0001). In addition, log K IPCE (SPC/Ch 80:20%) values correlated well with the pIC 50 (the minus logarithm of IC 50) values for cyclooxygenase 2 determined on intact cells ( R = 0.959 and P < 0.001). These results confirmed that IPCE, K IPCE value as evaluation index, can be effectually used for studying drug–membrane interactions and it has the potential to predict drug activity. Cholesterol-containing (20 mol%) liposomes may be more suitable to mimic real cell membrane.

MILES: Multiple-Instance Learning via Embedded Instance Selection

Multiple-instance problems arise from the situations where training class labels are attached to sets of samples (named bags), instead of individual samples within each bag (called instances). Most previous multiple-instance learning (MIL) algorithms are developed based on the assumption that a bag is positive if and only if at least one of its instances is positive. Although the assumption works well in a drug activity prediction problem, it is rather restrictive for other applications, especially those in the computer vision area. We propose a learning method, MILES (Multiple-Instance Learning via Embedded instance Selection), which converts the multiple-instance learning problem to a standard supervised learning problem that does not impose the assumption relating instance labels to bag labels. MILES maps each bag into a feature space defined by the instances in the training bags via an instance similarity measure. This feature mapping often provides a large number of redundant or irrelevant features. Hence, 1-norm SVM is applied to select important features as well as construct classifiers simultaneously. We have performed extensive experiments. In comparison with other methods, MILES demonstrates competitive classification accuracy, high computation efficiency, and robustness to labeling uncertainty.

Image Categorization by Learning and Reasoning with Regions

Designing computer programs to automatically categorize images using low-level features is a challenging research topic in computer vision. In this paper, we present a new learning technique, which extends Multiple-Instance Learning (MIL), and its application to the problem of region-based image categorization. Images are viewed as bags, each of which contains a number of instances corresponding to regions obtained from image segmentation. The standard MIL problem assumes that a bag is labeled positive if at least one of its instances is positive; otherwise, the bag is negative. In the proposed MIL framework, DD-SVM, a bag label is determined by some number of instances satisfying various properties. DD-SVM first learns a collection of instance prototypes according to a Diverse Density (DD) function. Each instance prototype represents a class of instances that is more likely to appear in bags with the specific label than in the other bags. A nonlinear mapping is then defined using the instance prototypes and maps every bag to a point in a new feature space, named the bag feature space. Finally, standard support vector machines are trained in the bag feature space. We provide experimental results on an image categorization problem and a drug activity prediction problem.

Quantitative Drug Activity Prediction for Inhibitors of Human Breast Carcinoma

Quantitative Drug Activity Prediction for Inhibitors of Human Breast Carcinoma

Quantitative Drug Activity Prediction for Inhibitors of Human Breast Carcinoma

There are sufficient data on a series of N−aryl−N-(2-chloroethyl) ureas tested as inhibitors of human breast carcinoma of the MDA-MB-231 cancer line to allow us to apply the so-called electron-conformational (EC) method in order to predict the drug activity of other molecular systems. The EC method reveals the pharmacophore and predicts the activity quantitatively based on conformational analysis and electronic structure calculations for a set of molecules that were tested for the activity under consideration (the training set). To reveal the pharmacophore of inhibitors of the human breast carcinoma MDA-MB-231 and to determine the parameters of molecular structure that enhance or diminish the activity, thus providing a tool for further increasing the inhibitor activity. Details of the EC method are described elsewhere. For the training set of the tested molecules, the electronic structure of the lowest conformations is calculated and presented in matrices, which are then processed in comparison with the activity. This reveals the pharmacophore: the special features of the drug molecules that are common for all active compounds and absent in the inactive ones. No arbitrary descriptors are used in the identification of the pharmacophore. The out-of-pharmacophore groups are then analysed and parameterised in an error-minimisation scheme that allows for calculation of the quantitative activity of any molecular system that has the pharmacophore. The pharmacophore of human breast carcinoma inhibitors is given by ten numbers with certain limits of flexibility. They represent four atoms (reactivity positions) with certain interatomic distances within the defined tolerances. Typical out-of-pharmacophore groups in the drug molecule serve as either antipharmacophore shields that diminish the activity or as other factors that may enhance the lipophilicity. A formula is obtained in which the roles of pharmacophore tolerances and out-of-pharmacophore group influences are presented by numerical coefficients, allowing the calculation of the expected activity quantitatively. The formula liability is estimated to be ≈97%. A cross-validation is performed showing that the prediction power is 92%. The pharmacophore evaluation is ≈100% correct, within the accuracy of the experimental data in the training set. An additional simple formula allows for the discrimination of the most active out-of-pharmacophore groups and the most influential pharmacophore tolerance parameters. The pharmacophore of the inhibitors of human breast carcinoma MDA-MB-231 is revealed and allows for high-accuracy data mining for potential drugs. A formula is derived for quantitative prediction of the activity of untested molecules with an expected accuracy of 92%.

Multiple instance learning of real valued data

The multiple-instance learning model has received much attention recently with a primary application area being that of drug activity prediction. Most prior work on multiple-instance learning has been for concept learning, yet for drug activity prediction, the label is a real-valued affinity measurement giving the binding strength. We present extensions of k-nearest neighbors (k-NN), Citation-kNN, and the diverse density algorithm for the real-valued setting and study their performance on Boolean and real-valued data. We also provide a method for generating chemically realistic artificial data.

The rationale for aerosolized antibiotics.

In order for an antimicrobial agent to be effective, it must fulfill two requirements. First, the agent must reach the site of infection and remain in the vicinity for an adequate length of time. Second, it must bind to a target site and remain bound for a length of time sufficient to disrupt the life cycle of the cell. Once these requirements are met, the drug is able to exert its antimicrobial activity against the cell. In an effort to better understand and predict the killing activity of antibiotics, we have attempted to develop parameters that describe the accumulation and diffusion of drug to and from body sites (pharmacokinetics) and quantify how much of a compound is needed at the site of infection to yield the desired effect (minimum inhibitory concentration). Furthermore, integration of these parameters allows us to evaluate host, drug, and microbial factors and formulate criteria to assess and predict drug activity in patients (pharmacodynamics). Knowledge and application of pharmacodynamic principles can assist clinicians in optimizing antimicrobial therapy by allowing them to maximize the antimicrobial activity of an agent while minimizing patient exposure and thus reducing the likelihood of toxicity.

Virulence of two strains of Cowdria ruminantium in mice and their use to predict drug activity against heartwater.

A study was made of the infectivity of two mouse-adapted strains of Cowdria ruminantium in mice. The Kwanyanga strain was most virulent in Balb/C mice which died nine days after infection with homogenate of liver from infected mice. CD-1 mice were least susceptible of six strains tested. The du Plessis strain of C. ruminantium was equally virulent in all six mouse strains. The du Plessis strain in CD-1 mice was used as the basis of a drug screen to detect activity against heartwater (C. ruminantium infection) and was highly predictive when active compounds were tested in sheep infected with the Ball 3 strain of C. ruminantium.