Information Loss in Network Pharmacology.

With the advent of increasing computational power and large-scale data acquisition, network analysis has become an attractive tool to study the organisation of complex systems and the interrelation of their constituent entities in various scientific domains. In many cases, relations only occur between entities of two different subsets, thereby forming a bipartite network. Often, the analysis of such bipartite networks involves the consideration of its two monopartite projections in order to focus on each entity subset individually as a means to deduce properties of the underlying original network. Although it is broadly acknowledged that this type of projection is not lossless, the inherent limitations of their interpretability are rarely discussed. In this work, we introduce two approaches for measuring the information loss associated with bipartite network projection. Application to two structurally distinct cases in network pharmacology, namely, drug-target and disease-gene bipartite networks, confirms that the major determinant of information loss is the degree of vertices omitted during the monopartite projection.

Systematic pharmacological screens uncover novel pathways involved in cerebral cavernous malformations.

Cerebral cavernous malformations (CCMs) are vascular lesions in the central nervous system causing strokes and seizures which currently can only be treated through neurosurgery. The disease arises through changes in the regulatory networks of endothelial cells that must be comprehensively understood to develop alternative, non-invasive pharmacological therapies. Here, we present the results of several unbiased small-molecule suppression screens in which we applied a total of 5,268 unique substances to CCM mutant worm, zebrafish, mouse, or human endothelial cells. We used a systems biology-based target prediction tool to integrate the results with the whole-transcriptome profile of zebrafish CCM2 mutants, revealing signaling pathways relevant to the disease and potential targets for small-molecule-based therapies. We found indirubin-3-monoxime to alleviate the lesion burden in murine preclinical models of CCM2 and CCM3 and suppress the loss-of-CCM phenotypes in human endothelial cells. Our multi-organism-based approach reveals new components of the CCM regulatory network and foreshadows novel small-molecule-based therapeutic applications for suppressing this devastating disease in patients.

A Novel Drug-Mouse Phenotypic Similarity Method Detects Molecular Determinants of Drug Effects.

The molecular mechanisms that translate drug treatment into beneficial and unwanted effects are largely unknown. We present here a novel approach to detect gene-drug and gene-side effect associations based on the phenotypic similarity of drugs and single gene perturbations in mice that account for the polypharmacological property of drugs. We scored the phenotypic similarity of human side effect profiles of 1,667 small molecules and biologicals to profiles of phenotypic traits of 5,384 mouse genes. The benchmarking with known relationships revealed a strong enrichment of physical and indirect drug-target connections, causative drug target-side effect links as well as gene-drug links involved in pharmacogenetic associations among phenotypically similar gene-drug pairs. The validation by in vitro assays and the experimental verification of an unknown connection between oxandrolone and prokineticin receptor 2 reinforces the ability of this method to provide new molecular insights underlying drug treatment. Thus, this approach may aid in the proposal of novel and personalized treatments.

Organ system heterogeneity DB: a database for the visualization of phenotypes at the organ system level.

Perturbations of mammalian organisms including diseases, drug treatments and gene perturbations in mice affect organ systems differently. Some perturbations impair relatively few organ systems while others lead to highly heterogeneous or systemic effects. Organ System Heterogeneity DB (http://mips.helmholtz-muenchen.de/Organ_System_Heterogeneity/) provides information on the phenotypic effects of 4865 human diseases, 1667 drugs and 5361 genetically modified mouse models on 26 different organ systems. Disease symptoms, drug side effects and mouse phenotypes are mapped to the System Organ Class (SOC) level of the Medical Dictionary of Regulatory Activities (MedDRA). Then, the organ system heterogeneity value, a measurement of the systemic impact of a perturbation, is calculated from the relative frequency of phenotypic features across all SOCs. For perturbations of interest, the database displays the distribution of phenotypic effects across organ systems along with the heterogeneity value and the distance between organ system distributions. In this way, it allows, in an easy and comprehensible fashion, the comparison of the phenotypic organ system distributions of diseases, drugs and their corresponding genetically modified mouse models of associated disease genes and drug targets. The Organ System Heterogeneity DB is thus a platform for the visualization and comparison of organ system level phenotypic effects of drugs, diseases and genes.

Molecularly and clinically related drugs and diseases are enriched in phenotypically similar drug-disease pairs.

BACKGROUND: The incomplete understanding of disease causes and drug mechanisms of action often leads to ineffective drug therapies or side effects. Therefore, new approaches are needed to improve treatment decisions and to elucidate molecular mechanisms underlying pathologies and unwanted drug effects. METHODS: We present here the first analysis of phenotypically related drug-disease pairs. The phenotypic similarity between 4,869 human diseases and 1,667 drugs was evaluated using an ontology-based semantic similarity approach to compare disease symptoms with drug side effects. We assessed and visualized the enrichment over random of clinical and molecular relationships among drug-disease pairs that share phenotypes using lift plots. To determine the associations between drug and disease classes enriched among phenotypically related pairs we employed a network-based approach combined with Fisher's exact test. RESULTS: We observed that molecularly and clinically related (for example, indication or contraindication) drugs and diseases are likely to share phenotypes. An analysis of the relations between drug mechanisms of action (MoAs) and disease classes among highly similar pairs revealed known and suspected MoA-disease relationships. Interestingly, we found that contraindications associated with high phenotypic similarity often involve diseases that have been reported as side effects of the drug, probably due to common mechanisms. Based on this, we propose a list of 752 precautions or potential contraindications for 486 drugs. CONCLUSIONS: Phenotypic similarity between drugs and diseases facilitates the proposal of contraindications and the mechanistic understanding of diseases and drug side effects.

Systematic analysis of gene properties influencing organ system phenotypes in mammalian perturbations.

MOTIVATION: Diseases and adverse drug reactions are frequently caused by disruptions in gene functionality. Gaining insight into the global system properties governing the relationships between genotype and phenotype is thus crucial to understand and interfere with perturbations in complex organisms such as diseases states. RESULTS: We present a systematic analysis of phenotypic information of 5047 perturbations of single genes in mice, 4766 human diseases and 1666 drugs that examines the relationships between different gene properties and the phenotypic impact at the organ system level in mammalian organisms. We observe that while single gene perturbations and alterations of nonessential, tissue-specific genes or those with low betweenness centrality in protein-protein interaction networks often show organ-specific effects, multiple gene alterations resulting e.g. from complex disorders and drug treatments have a more widespread impact. Interestingly, certain cellular localizations are distinctly associated to systemic effects in monogenic disease genes and mouse gene perturbations, such as the lumen of intracellular organelles and transcription factor complexes, respectively. In summary, we show that the broadness of the phenotypic effect is clearly related to certain gene properties and is an indicator of the severity of perturbations. This work contributes to the understanding of gene properties influencing the systemic effects of diseases and drugs.

HitPick: a web server for hit identification and target prediction of chemical screenings.

MOTIVATION: High-throughput phenotypic assays reveal information about the molecules that modulate biological processes, such as a disease phenotype and a signaling pathway. In these assays, the identification of hits along with their molecular targets is critical to understand the chemical activities modulating the biological system. Here, we present HitPick, a web server for identification of hits in high-throughput chemical screenings and prediction of their molecular targets. HitPick applies the B-score method for hit identification and a newly developed approach combining 1-nearest-neighbor (1NN) similarity searching and Laplacian-modified naïve Bayesian target models to predict targets of identified hits. The performance of the HitPick web server is presented and discussed. AVAILABILITY: The server can be accessed at http://mips.helmholtz-muenchen.de/proj/hitpick. CONTACT: monica.campillos@helmholtz-muenchen.de.

Drug-Target Networks.

Far from the traditional view of selective drug-target interactions, the recent accumulation of large amounts of interaction data for small-molecule drugs and protein targets requires innovative visualisation and analysis tools that are able to deal with what has become a truly complex system. In this context, network theory offers both a robust and illustrative framework to investigate drug-target connections and has been swiftly and widely embraced by the chemical biology and molecular informatics communities. A survey of the most recent applications of drug-target networks to detect cross-pharmacology relationships among targets and to identify new targets for known drugs is provided. Finally, some of the current limitations are also discussed, including the actual completeness of interaction data and the information loss intrinsically associated with the one-mode projection of drug-target networks.

Design and exploration of target-selective chemical space representations.

We report the design of target-selective chemical spaces using CA-DynaMAD, a mapping algorithm that generates and navigates flexible space representations for the identification of active or selective compounds. The algorithm iteratively increases the dimensionality of reference spaces in a controlled manner by evaluating a single descriptor per iteration. For seven sets of closely related biogenic amine G protein coupled receptor (GPCR) antagonists with different selectivity, target-selective reference spaces were designed and used to identify selective compounds by screening a biologically annotated database. Combinations of descriptors that constitute target-selective reference spaces identified with CA-DynaMAD can also be used to build other computational models for the prediction of compound selectivity.

Exploring structure-selectivity relationships of biogenic amine GPCR antagonists using similarity searching and dynamic compound mapping.

We design and analyze compound selectivity sets of antagonists with differential selectivity against seven biogenic amine G-protein coupled receptors. The selectivity sets consist of a total of 267 antagonists and contain a spectrum of in part closely related molecular scaffolds. Each set represents a different selectivity profile. Using these com- pound sets, a systematic computational analysis of structure-selectivity relationships is carried out with different 2D similarity methods including fingerprints, recursive partitioning, clustering, and dynamic compound mapping. Screening calculations are performed in a background database containing nearly four million molecules. Fingerprint searching and compound mapping are found to enrich target-selective antagonists over family-selective ones. Dynamic compound mapping effectively discriminates database compounds from GPCR antagonists and consistently retains target-selective antagonists during the final dimension extension levels. Furthermore, the widely used MACCS key fingerprint displays a strong tendency to distinguish between target- and family-selective GPCR antagonists. Taken together, the results indicate that different types of 2D similarity methods are capable of distinguishing closely related molecules having different selectivity. The reported compound benchmark system is made freely available in order to enable selectivity-oriented analyses using other computational approaches.

Methods for computer-aided chemical biology. Part 1: Design of a benchmark system for the evaluation of compound selectivity.

Computational drug design and discovery methods have traditionally put much emphasis on the identification of novel active compounds and the optimization of their potency. For chemical genetics and genomics applications, an important task is the identification of small molecules that are selective against target families, subfamilies, or individual targets and can be used as molecular probes for specific functions. In order to develop or tune computational methods for such applications, there is a need for molecular benchmark systems that focus on compound selectivity, rather than biological activity (in qualitative terms) or potency. We have constructed a selectivity-oriented test system that consists of 26 compound selectivity sets against 13 individual targets belonging to three distinct families and contains a total of 558 selective compounds. The targets were chosen because of pharmaceutical relevance and the availability of suitable ligands, privileged structural motifs and/or target structure information. Compound selectivity sets were characterized by structural diversity, chemical scaffold and selectivity range analysis. The test system is made freely available and should be useful for the development of computational approaches in chemical biology.

Methods for computer-aided chemical biology. Part 2: Evaluation of compound selectivity using 2D molecular fingerprints.

We analyze 558 compounds with selectivity against members of different protein families using two-dimensional molecular fingerprint methods. The calculations target compounds selective for 13 targets belonging to three families. These compound sets were especially designed for selectivity studies. The identification of compounds displaying different selectivity patterns against related protein targets is a prerequisite for chemical genetics and genomics applications to specifically interfere with functions of individual members of protein families. Thus far, computational methods have only little impact on the search for selective compounds. This is in part due to the fact that selectivity is more difficult to study computationally than activity because selectivity analysis requires the evaluation of compounds binding to multiple targets. Here, we investigate the ability of state-of-the-art two-dimensional molecular fingerprints to detect compounds having different selectivity. The results of systematic similarity search calculations reveal that two-dimensional fingerprints are capable of identifying compounds having different selectivity against closely related target proteins, although fingerprints were originally not developed for such applications. In addition to target-selective molecules, fingerprints are also found to preferentially recognize compounds that are active at the target family level. Our findings suggest that similarity methods should merit further exploration in the study of compound selectivity across target families.

Analysis of a high-throughput screening data set using potency-scaled molecular similarity algorithms.

Molecular similarity methods for ligand-based virtual screening (VS) generally do not take compound potency as a variable or search parameter into account. We have incorporated a logarithmic potency scaling function into two conceptually distinct VS algorithms to account for relative compound potency during search calculations. A high-throughput screening (HTS) data set containing cathepsin B inhibitors was analyzed to evaluate the effects of potency scaling. Sets of template compounds were randomly selected from the HTS data and used to search for hits having varying potency levels in the presence or absence of potency scaling. Enrichment of potent compounds in small subsets of the HTS data set was observed as a consequence of potency scaling. In part, observed enrichments could be rationalized as a result of recentering chemical reference space on a subspace populated by potent compounds. Our findings suggest that VS calculations using multiple reference compounds can be directed toward the preferential detection of potent database hits by scaling compound contributions according to potency differences.

Mapping algorithms for molecular similarity analysis and ligand-based virtual screening: design of DynaMAD and comparison with MAD and DMC.

Here, we introduce the DynaMAD algorithm that is designed to map database compounds to combinations of activity-class-dependent descriptor value ranges in order to identify novel active molecules. The method combines and extends key features of two previously developed algorithms, MAD and DMC. These methods were first described as compound-mapping algorithms for large-scale virtual screening applications. DynaMAD and DMC operate in chemical spaces of stepwise increasing dimensionality. However, in contrast to DMC, which utilizes binary transformed descriptors, DynaMAD uses unmodified descriptor value distributions. The performance of these mapping methods was compared in detail in virtual screening trials on 24 different compound activity classes against a background of about 2 million database compounds. In these calculations, all three approaches produced results of considerable predictive value, and the enrichment of active molecules in small selection sets consisting of only about 20 or fewer database compounds emerged as a common feature. Furthermore, mapping methods were capable of recognizing remote molecular similarity relationships. Overall, DynaMAD performed better than MAD and DMC, producing average hit and recovery rates of 55% and 33%, respectively, over all 24 classes. Taken together, our findings suggest that dynamic compound mapping to combinations of activity-class-selective descriptor settings has significant potential for molecular similarity analysis and ligand-based virtual screening.