NEMix: Single-cell Nested Effects Models for Probabilistic Pathway Stimulation

Nested effects models have been used successfully for learning subcellular networks from high-dimensional perturbation effects that result from RNA interference (RNAi) experiments. Here, we further develop the basic nested effects model using high-content single-cell imaging data from RNAi screens of cultured cells infected with human rhinovirus. RNAi screens with single-cell readouts are becoming increasingly common, and they often reveal high cell-to-cell variation. As a consequence of this cellular heterogeneity, knock-downs result in variable effects among cells and lead to weak average phenotypes on the cell population level. To address this confounding factor in network inference, we explicitly model the stimulation status of a signaling pathway in individual cells. We extend the framework of nested effects models to probabilistic combinatorial knock-downs and propose NEMix, a nested effects mixture model that accounts for unobserved pathway activation. We analyzed the identifiability of NEMix and developed a parameter inference scheme based on the Expectation Maximization algorithm. In an extensive simulation study, we show that NEMix improves learning of pathway structures over classical NEMs significantly in the presence of hidden pathway stimulation. We applied our model to single-cell imaging data from RNAi screens monitoring human rhinovirus infection, where limited infection efficiency of the assay results in uncertain pathway stimulation. Using a subset of genes with known interactions, we show that the inferred NEMix network has high accuracy and outperforms the classical nested effects model without hidden pathway activity. NEMix is implemented as part of the R/Bioconductor package ‘nem’ and available at www.cbg.ethz.ch/software/NEMix.     

Structure learning in Nested Effects Models

Nested Effects Models (NEMs) are a class of graphical models introduced to analyze the results of gene perturbation screens. NEMs explore noisy subset relations between the high-dimensional outputs of phenotyping studies, e.g., the effects showing in gene expression profiles or as morphological features of the perturbed cell. In this paper we expand the statistical basis of NEMs in four directions. First, we derive a new formula for the likelihood function of a NEM, which generalizes previous results for binary data. Second, we prove model identifiability under mild assumptions. Third, we show that the new formulation of the likelihood allows efficiency in traversing model space. Fourth, we incorporate prior knowledge and an automated variable selection criterion to decrease the influence of noise in the data.     

A Bayesian Network View on Nested Effects Models

Nested effects models (NEMs) are a class of probabilistic models that were designed to reconstruct a hidden signalling structure from a large set of observable effects caused by active interventions into the signalling pathway. We give a more flexible formulation of NEMs in the language of Bayesian networks. Our framework constitutes a natural generalization of the original NEM model, since it explicitly states the assumptions that are tacitly underlying the original version. Our approach gives rise to new learning methods for NEMs, which have been implemented in the /Bioconductor package nem. We validate these methods in a simulation study and apply them to a synthetic lethality dataset in yeast.     

Nested effects models for high-dimensional phenotyping screens

MOTIVATION: In high-dimensional phenotyping screens, a large number of cellular features is observed after perturbing genes by knockouts or RNA interference. Comprehensive analysis of perturbation effects is one of the most powerful techniques for attributing functions to genes, but not much work has been done so far to adapt statistical and computational methodology to the specific needs of large-scale and high-dimensional phenotyping screens. RESULTS: We introduce and compare probabilistic methods to efficiently infer a genetic hierarchy from the nested structure of observed perturbation effects. These hierarchies elucidate the structures of signaling pathways and regulatory networks. Our methods achieve two goals: (1) they reveal clusters of genes with highly similar phenotypic profiles, and (2) they order (clusters of) genes according to subset relationships between phenotypes. We evaluate our algorithms in the controlled setting of simulation studies and show their practical use in two experimental scenarios: (1) a data set investigating the response to microbial challenge in Drosophila melanogaster, and (2) a compendium of expression profiles of Saccharomyces cerevisiae knockout strains. We show that our methods identify biologically justified genetic hierarchies of perturbation effects. AVAILABILITY: The software used in our analysis is freely available in the R package 'nem' from www.bioconductor.org.     

Analysis of sample set enrichment scores: assaying the enrichment of sets of genes for individual samples in genome-wide expression profiles

MOTIVATION: Gene expression profiling experiments in cell lines and animal models characterized by specific genetic or molecular perturbations have yielded sets of genes annotated by the perturbation. These gene sets can serve as a reference base for interrogating other expression datasets. For example, a new dataset in which a specific pathway gene set appears to be enriched, in terms of multiple genes in that set evidencing expression changes, can then be annotated by that reference pathway. We introduce in this paper a formal statistical method to measure the enrichment of each sample in an expression dataset. This allows us to assay the natural variation of pathway activity in observed gene expression data sets from clinical cancer and other studies. RESULTS: Validation of the method and illustrations of biological insights gleaned are demonstrated on cell line data, mouse models, and cancer-related datasets. Using oncogenic pathway signatures, we show that gene sets built from a model system are indeed enriched in the model system. We employ ASSESS for the use of molecular classification by pathways. This provides an accurate classifier that can be interpreted at the level of pathways instead of individual genes. Finally, ASSESS can be used for cross-platform expression models where data on the same type of cancer are integrated over different platforms into a space of enrichment scores. AVAILABILITY: Versions are available in Octave and Java (with a graphical user interface). Software can be downloaded at http://people.genome.duke.edu/assess.     

A Structural Systems Biology Approach for Quantifying the Systemic Consequences of Missense Mutations in Proteins

Gauging the systemic effects of non-synonymous single nucleotide polymorphisms (nsSNPs) is an important topic in the pursuit of personalized medicine. However, it is a non-trivial task to understand how a change at the protein structure level eventually affects a cell''s behavior. This is because complex information at both the protein and pathway level has to be integrated. Given that the idea of integrating both protein and pathway dynamics to estimate the systemic impact of missense mutations in proteins remains predominantly unexplored, we investigate the practicality of such an approach by formulating mathematical models and comparing them with experimental data to study missense mutations. We present two case studies: (1) interpreting systemic perturbation for mutations within the cell cycle control mechanisms (G2 to mitosis transition) for yeast; (2) phenotypic classification of neuron-related human diseases associated with mutations within the mitogen-activated protein kinase (MAPK) pathway. We show that the application of simplified mathematical models is feasible for understanding the effects of small sequence changes on cellular behavior. Furthermore, we show that the systemic impact of missense mutations can be effectively quantified as a combination of protein stability change and pathway perturbation.     

BioPAX support in CellDesigner

MOTIVATION: BioPAX is a standard language for representing and exchanging models of biological processes at the molecular and cellular levels. It is widely used by different pathway databases and genomics data analysis software. Currently, the primary source of BioPAX data is direct exports from the curated pathway databases. It is still uncommon for wet-lab biologists to share and exchange pathway knowledge using BioPAX. Instead, pathways are usually represented as informal diagrams in the literature. In order to encourage formal representation of pathways, we describe a software package that allows users to create pathway diagrams using CellDesigner, a user-friendly graphical pathway-editing tool and save the pathway data in BioPAX Level 3 format. AVAILABILITY: The plug-in is freely available and can be downloaded at ftp://ftp.pantherdb.org/CellDesigner/plugins/BioPAX/ CONTACT: huaiyumi@usc.edu SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.     

Pathway Correlation Profile of Gene-Gene Co-Expression for Identifying Pathway Perturbation

Identifying perturbed or dysregulated pathways is critical to understanding the biological processes that change within an experiment. Previous methods identified important pathways that are significantly enriched among differentially expressed genes; however, these methods cannot account for small, coordinated changes in gene expression that amass across a whole pathway. In order to overcome this limitation, we use microarray gene expression data to identify pathway perturbation based on pathway correlation profiles. By identifying the distribution of gene-gene pair correlations within a pathway, we can rank the pathways based on the level of perturbation and dysregulation. We have shown this successfully for differences between two experimental conditions in Escherichia coli and changes within time series data in Saccharomyces cerevisiae, as well as two estrogen receptor response classes of breast cancer. Overall, our method made significant predictions as to the pathway perturbations that are involved in the experimental conditions.     

Predicting phenotypic effects of gene perturbations in C. elegans using an integrated network model

Predicting the phenotype of an organism from its genotype is a central question in genetics. Most importantly, we would like to find out if the perturbation of a single gene may be the cause of a disease. However, our current ability to predict the phenotypic effects of perturbations of individual genes is limited. Network models of genes are one tool for tackling this problem. In a recent study, (Lee et al.) it has been shown that network models covering the majority of genes of an organism can be used for accurately predicting phenotypic effects of gene perturbations in multicellular organisms. .     

Inflammatory aetiology of human myometrial activation tested using directed graphs

There are three main hypotheses for the activation of the human uterus at labour: functional progesterone withdrawal, inflammatory stimulation, and oxytocin receptor activation. To test these alternatives we have taken information and data from the literature to develop causal pathway models for the activation of human myometrium. The data provided quantitative RT-PCR results on key genes from samples taken before and during labour. Principal component analysis showed that pre-labour samples form a homogenous group compared to those during labour. We therefore modelled the alternative causal pathways in non-laboring samples using directed graphs and statistically compared the likelihood of the different models using structural equations and D-separation approaches. Using the computer program LISREL, inflammatory activation as a primary event was highly consistent with the data (p = 0.925), progesterone withdrawal, as a primary event, is plausible (p = 0.499), yet comparatively unlikely, oxytocin receptor mediated initiation is less compatible with the data (p = 0.091). DGraph, a software program that creates directed graphs, produced similar results (p= 0.684, p= 0.280, and p = 0.04, respectively). This outcome supports an inflammatory aetiology for human labour. Our results demonstrate the value of directed graphs in determining the likelihood of causal relationships in biology in situations where experiments are not possible.     

Improved Reconstruction of In Silico Gene Regulatory Networks by Integrating Knockout and Perturbation Data

We performed computational reconstruction of the in silico gene regulatory networks in the DREAM3 Challenges. Our task was to learn the networks from two types of data, namely gene expression profiles in deletion strains (the ‘deletion data’) and time series trajectories of gene expression after some initial perturbation (the ‘perturbation data’). In the course of developing the prediction method, we observed that the two types of data contained different and complementary information about the underlying network. In particular, deletion data allow for the detection of direct regulatory activities with strong responses upon the deletion of the regulator while perturbation data provide richer information for the identification of weaker and more complex types of regulation. We applied different techniques to learn the regulation from the two types of data. For deletion data, we learned a noise model to distinguish real signals from random fluctuations using an iterative method. For perturbation data, we used differential equations to model the change of expression levels of a gene along the trajectories due to the regulation of other genes. We tried different models, and combined their predictions. The final predictions were obtained by merging the results from the two types of data. A comparison with the actual regulatory networks suggests that our approach is effective for networks with a range of different sizes. The success of the approach demonstrates the importance of integrating heterogeneous data in network reconstruction.     

Developmental instability of the Drosophila wing as an index of genomic perturbation and altered cell proliferation

We experimentally induced different levels of instability affecting the development of specific wing regions of Drosophila melanogaster using the UAS-GAL4 system. A common index of developmental instability is fluctuating asymmetry (FA), that is, random differences between body sides of single individuals. We studied the FA in transgenic strains carrying random genomic insertions (UAS strains), as well as insertions in the regulatory region of genes involved in the organization of wing development (GAL4 strains). In addition, the expression of genes that increase (dp110 and 3622) or decrease (dPTEN) cell proliferation was ectopically induced. Our results are related to different levels of perturbation. Through the first kind of perturbation, genome integrity was compromised by the insertion of foreign DNA. In all cases, we observed a general increase in FA, although it was rarely found significant. The second kind of perturbation involved a modification of genes controlling wing development through the insertion of a GAL4 sequence in their promoter region. The third kind involved the ectopic expression of genes controlling cell proliferation. Our results show that (i) the level of FA is connected with the level of morphological perturbation induced, (ii) FA increase was higher in the wing regions that were the target of the genetic perturbation, and (iii) developmental instability was also observed in regions that were not directly addressed by the perturbation. The results were discussed on the basis of the running models about Drosophila wing development.     

Discovery of Drug Synergies in Gastric Cancer Cells Predicted by Logical Modeling

Discovery of efficient anti-cancer drug combinations is a major challenge, since experimental testing of all possible combinations is clearly impossible. Recent efforts to computationally predict drug combination responses retain this experimental search space, as model definitions typically rely on extensive drug perturbation data. We developed a dynamical model representing a cell fate decision network in the AGS gastric cancer cell line, relying on background knowledge extracted from literature and databases. We defined a set of logical equations recapitulating AGS data observed in cells in their baseline proliferative state. Using the modeling software GINsim, model reduction and simulation compression techniques were applied to cope with the vast state space of large logical models and enable simulations of pairwise applications of specific signaling inhibitory chemical substances. Our simulations predicted synergistic growth inhibitory action of five combinations from a total of 21 possible pairs. Four of the predicted synergies were confirmed in AGS cell growth real-time assays, including known effects of combined MEK-AKT or MEK-PI3K inhibitions, along with novel synergistic effects of combined TAK1-AKT or TAK1-PI3K inhibitions. Our strategy reduces the dependence on a priori drug perturbation experimentation for well-characterized signaling networks, by demonstrating that a model predictive of combinatorial drug effects can be inferred from background knowledge on unperturbed and proliferating cancer cells. Our modeling approach can thus contribute to preclinical discovery of efficient anticancer drug combinations, and thereby to development of strategies to tailor treatment to individual cancer patients.     

Inferring the perturbed microRNA regulatory networks from gene expression data using a network propagation based method

Background

MicroRNAs (miRNAs) are a class of endogenous small regulatory RNAs. Identifications of the dys-regulated or perturbed miRNAs and their key target genes are important for understanding the regulatory networks associated with the studied cellular processes. Several computational methods have been developed to infer the perturbed miRNA regulatory networks by integrating genome-wide gene expression data and sequence-based miRNA-target predictions. However, most of them only use the expression information of the miRNA direct targets, rarely considering the secondary effects of miRNA perturbation on the global gene regulatory networks.

Results

We proposed a network propagation based method to infer the perturbed miRNAs and their key target genes by integrating gene expressions and global gene regulatory network information. The method used random walk with restart in gene regulatory networks to model the network effects of the miRNA perturbation. Then, it evaluated the significance of the correlation between the network effects of the miRNA perturbation and the gene differential expression levels with a forward searching strategy. Results show that our method outperformed several compared methods in rediscovering the experimentally perturbed miRNAs in cancer cell lines. Then, we applied it on a gene expression dataset of colorectal cancer clinical patient samples and inferred the perturbed miRNA regulatory networks of colorectal cancer, including several known oncogenic or tumor-suppressive miRNAs, such as miR-17, miR-26 and miR-145.

Conclusions

Our network propagation based method takes advantage of the network effect of the miRNA perturbation on its target genes. It is a useful approach to infer the perturbed miRNAs and their key target genes associated with the studied biological processes using gene expression data.

Electronic supplementary material

The online version of this article (doi:10.1186/1471-2105-15-255) contains supplementary material, which is available to authorized users.