Learning signaling network structures with sparsely distributed data

Data-driven learning how oncogenic gene expression locally alters heterocellular networks

Developing drugs increasingly relies on mechanistic modeling and simulation. Models that capture causal relations among genetic drivers of oncogenesis, functional plasticity, and host immunity complement wet experiments. Unfortunately, formulating such mechanistic cell-level models currently relies on hand curation, which can bias how data is interpreted or the priority of drug targets. In modeling molecular-level networks, rules and algorithms are employed to limit a priori biases in formulating mechanistic models. Here we combine digital cytometry with Bayesian network inference to generate causal models of cell-level networks linking an increase in gene expression associated with oncogenesis with alterations in stromal and immune cell subsets from bulk transcriptomic datasets. We predict how increased Cell Communication Network factor 4, a secreted matricellular protein, alters the tumor microenvironment using data from patients diagnosed with breast cancer and melanoma. Predictions are then tested using two immunocompetent mouse models for melanoma, which provide consistent experimental results.

While mechanistic models play increasing roles in immuno-oncology, hand network curation is current practice. Here the authors use a Bayesian data-driven approach to infer how expression of a secreted oncogene alters the cellular landscape within the tumor.

Identifying causal networks linking cancer processes and anti-tumor immunity using Bayesian network inference and metagene constructs

Cancer arises from a deregulation of both intracellular and intercellular networks that maintain system homeostasis. Identifying the architecture of these networks and how they are changed in cancer is a pre-requisite for designing drugs to restore homeostasis. Since intercellular networks only appear in intact systems, it is difficult to identify how these networks become altered in human cancer using many of the common experimental models. To overcome this, we used the diversity in normal and malignant human tissue samples from the Cancer Genome Atlas (TCGA) database of human breast cancer to identify the topology associated with intercellular networks in vivo. To improve the underlying biological signals, we constructed Bayesian networks using metagene constructs, which represented groups of genes that are concomitantly associated with different immune and cancer states. We also used bootstrap resampling to establish the significance associated with the inferred networks. In short, we found opposing relationships between cell proliferation and epithelial-to-mesenchymal transformation (EMT) with regards to macrophage polarization. These results were consistent across multiple carcinomas in that proliferation was associated with a type 1 cell-mediated anti-tumor immune response and EMT was associated with a pro-tumor anti-inflammatory response. To address the identifiability of these networks from other datasets, we could identify the relationship between EMT and macrophage polarization with fewer samples when the Bayesian network was generated from malignant samples alone. However, the relationship between proliferation and macrophage polarization was identified with fewer samples when the samples were taken from a combination of the normal and malignant samples. © 2016 American Institute of Chemical Engineers Biotechnol. Prog., 32:470-479, 2016.

Uncovering distinct protein-network topologies in heterogeneous cell populations

Background

Cell biology research is fundamentally limited by the number of intracellular components, particularly proteins, that can be co-measured in the same cell. Therefore, cell-to-cell heterogeneity in unmeasured proteins can lead to completely different observed relations between the same measured proteins. Attempts to infer such relations in a heterogeneous cell population can yield uninformative average relations if only one underlying biochemical network is assumed. To address this, we developed a method that recursively couples an iterative unmixing process with a Bayesian analysis of each unmixed subpopulation.

Results

Our approach enables to identify the number of distinct cell subpopulations, unmix their corresponding observations and resolve the network structure of each subpopulation. Using simulations of the MAPK pathway upon EGF and NGF stimulations we assess the performance of the method. We demonstrate that the presented method can identify better than clustering approaches the number of subpopulations within a mixture of observations, thus resolving correctly the statistical relations between the proteins.

Conclusions

Coupling the unmixing of multiplexed observations with the inference of statistical relations between the measured parameters is essential for the success of both of these processes. Here we present a conceptual and algorithmic solution to achieve such coupling and hence to analyze data obtained from a natural mixture of cell populations. As the technologies and necessity for multiplexed measurements are rising in the systems biology era, this work addresses an important current challenge in the analysis of the derived data.

Electronic supplementary material

The online version of this article (doi:10.1186/s12918-015-0170-2) contains supplementary material, which is available to authorized users.

Summary of the DREAM8 Parameter Estimation Challenge: Toward Parameter Identification for Whole-Cell Models

Whole-cell models that explicitly represent all cellular components at the molecular level have the potential to predict phenotype from genotype. However, even for simple bacteria, whole-cell models will contain thousands of parameters, many of which are poorly characterized or unknown. New algorithms are needed to estimate these parameters and enable researchers to build increasingly comprehensive models. We organized the Dialogue for Reverse Engineering Assessments and Methods (DREAM) 8 Whole-Cell Parameter Estimation Challenge to develop new parameter estimation algorithms for whole-cell models. We asked participants to identify a subset of parameters of a whole-cell model given the model’s structure and in silico “experimental” data. Here we describe the challenge, the best performing methods, and new insights into the identifiability of whole-cell models. We also describe several valuable lessons we learned toward improving future challenges. Going forward, we believe that collaborative efforts supported by inexpensive cloud computing have the potential to solve whole-cell model parameter estimation.

Whole-cell models promise to enable rational bioengineering by predicting how cells behave. Even for simple bacteria, whole-cell models require thousands of parameters, many of which are poorly characterized or unknown. New approaches are needed to estimate these parameters. We organized the Dialogue for Reverse Engineering Assessments and Methods (DREAM) 8 Whole-Cell Parameter Estimation Challenge to develop new approaches for whole-cell model parameter identification. Here we describe the challenge, the best performing methods, new insights into the identifiability of whole-cell models, and several lessons we learned for improving future challenges. Going forward, we believe that collaborative efforts have the potential to produce powerful tools for identifying whole-cell models.

Efficient Bayesian estimates for discrimination among topologically different systems biology models

A major effort in systems biology is the development of mathematical models that describe complex biological systems at multiple scales and levels of abstraction. Determining the topology-the set of interactions-of a biological system from observations of the system's behavior is an important and difficult problem. Here we present and demonstrate new methodology for efficiently computing the probability distribution over a set of topologies based on consistency with existing measurements. Key features of the new approach include derivation in a Bayesian framework, incorporation of prior probability distributions of topologies and parameters, and use of an analytically integrable linearization based on the Fisher information matrix that is responsible for large gains in efficiency. The new method was demonstrated on a collection of four biological topologies representing a kinase and phosphatase that operate in opposition to each other with either processive or distributive kinetics, giving 8-12 parameters for each topology. The linearization produced an approximate result very rapidly (CPU minutes) that was highly accurate on its own, as compared to a Monte Carlo method guaranteed to converge to the correct answer but at greater cost (CPU weeks). The Monte Carlo method developed and applied here used the linearization method as a starting point and importance sampling to approach the Bayesian answer in acceptable time. Other inexpensive methods to estimate probabilities produced poor approximations for this system, with likelihood estimation showing its well-known bias toward topologies with more parameters and the Akaike and Schwarz Information Criteria showing a strong bias toward topologies with fewer parameters. These results suggest that this linear approximation may be an effective compromise, providing an answer whose accuracy is near the true Bayesian answer, but at a cost near the common heuristics.

Single timepoint models of dynamic systems

Many interesting studies aimed at elucidating the connectivity structure of biomolecular pathways make use of abundance measurements, and employ statistical and information theoretic approaches to assess connectivities. These studies often do not address the effects of the dynamics of the underlying biological system, yet dynamics give rise to impactful issues such as timepoint selection and its effect on structure recovery. In this work, we study conditions for reliable retrieval of the connectivity structure of a dynamic system, and the impact of dynamics on structure-learning efforts. We encounter an unexpected problem not previously described in elucidating connectivity structure from dynamic systems, show how this confounds structure learning of the system and discuss possible approaches to overcome the confounding effect. Finally, we test our hypotheses on an accurate dynamic model of the IGF signalling pathway. We use two structure-learning methods at four time points to contrast the performance and robustness of those methods in terms of recovering correct connectivity.

Reconstruction of cellular signal transduction networks using perturbation assays and linear programming

Perturbation experiments for example using RNA interference (RNAi) offer an attractive way to elucidate gene function in a high throughput fashion. The placement of hit genes in their functional context and the inference of underlying networks from such data, however, are challenging tasks. One of the problems in network inference is the exponential number of possible network topologies for a given number of genes. Here, we introduce a novel mathematical approach to address this question. We formulate network inference as a linear optimization problem, which can be solved efficiently even for large-scale systems. We use simulated data to evaluate our approach, and show improved performance in particular on larger networks over state-of-the art methods. We achieve increased sensitivity and specificity, as well as a significant reduction in computing time. Furthermore, we show superior performance on noisy data. We then apply our approach to study the intracellular signaling of human primary nave CD4⁺ T-cells, as well as ErbB signaling in trastuzumab resistant breast cancer cells. In both cases, our approach recovers known interactions and points to additional relevant processes. In ErbB signaling, our results predict an important role of negative and positive feedback in controlling the cell cycle progression.

Measurement and modeling of signaling at the single-cell level

It has long been recognized that a deeper understanding of cell function, with respect to execution of phenotypic behaviors and their regulation by the extracellular environment, is likely to be achieved by analyzing the underlying molecular processes for individual cells selected from across a population, rather than averages of many cells comprising that population. In recent years, experimental and computational methods for undertaking these analyses have advanced rapidly. In this review, we provide a perspective on both measurement and modeling facets of biochemistry at a single-cell level. Our central focus is on receptor-mediated signaling networks that regulate cell phenotypic functions.

Computational analysis of high-throughput flow cytometry data

INTRODUCTION

Flow cytometry has been around for over 40 years, but only recently has the opportunity arisen to move into the high-throughput domain. The technology is now available and is highly competitive with imaging tools under the right conditions. Flow cytometry has, however, been a technology that has focused on its unique ability to study single cells and appropriate analytical tools are readily available to handle this traditional role of the technology.

AREAS COVERED

Expansion of flow cytometry to a high-throughput (HT) and high-content technology requires both advances in hardware and analytical tools. The historical perspective of flow cytometry operation as well as how the field has changed and what the key changes have been discussed. The authors provide a background and compelling arguments for moving toward HT flow, where there are many innovative opportunities. With alternative approaches now available for flow cytometry, there will be a considerable number of new applications. These opportunities show strong capability for drug screening and functional studies with cells in suspension.

EXPERT OPINION

There is no doubt that HT flow is a rich technology awaiting acceptance by the pharmaceutical community. It can provide a powerful phenotypic analytical toolset that has the capacity to change many current approaches to HT screening. The previous restrictions on the technology, based on its reduced capacity for sample throughput, are no longer a major issue. Overcoming this barrier has transformed a mature technology into one that can focus on systems biology questions not previously considered possible.

From single cells to deep phenotypes in cancer

In recent years, major advances in single-cell measurement systems have included the introduction of high-throughput versions of traditional flow cytometry that are now capable of measuring intracellular network activity, the emergence of isotope labels that can enable the tracking of a greater variety of cell markers and the development of super-resolution microscopy techniques that allow measurement of RNA expression in single living cells. These technologies will facilitate our capacity to catalog and bring order to the inherent diversity present in cancer cell populations. Alongside these developments, new computational approaches that mine deep data sets are facilitating the visualization of the shape of the data and enabling the extraction of meaningful outputs. These applications have the potential to reveal new insights into cancer biology at the intersections of stem cell function, tumor-initiating cells and multilineage tumor development. In the clinic, they may also prove important not only in the development of new diagnostic modalities but also in understanding how the emergence of tumor cell clones harboring different sets of mutations predispose patients to relapse or disease progression.

Imaging the coordination of multiple signalling activities in living cells

Cellular signal transduction occurs in complex and redundant interaction networks, which are best understood by simultaneously monitoring the activation dynamics of multiple components. Recent advances in biosensor technology have made it possible to visualize and quantify the activation of multiple network nodes in the same living cell. The precision and scope of this approach has been greatly extended by novel computational approaches (referred to as computational multiplexing) that can reveal relationships between network nodes imaged in separate cells.

Principles and strategies for developing network models in cancer

The flood of genome-wide data generated by high-throughput technologies currently provides biologists with an unprecedented opportunity: to manipulate, query, and reconstruct functional molecular networks of cells. Here, we outline three underlying principles and six strategies to infer network models from genomic data. Then, using cancer as an example, we describe experimental and computational approaches to infer "differential" networks that can identify genes and processes driving disease phenotypes. In conclusion, we discuss how a network-level understanding of cancer can be used to predict drug response and guide therapeutics.

A comprehensive statistical model for cell signaling

Protein signaling networks play a central role in transcriptional regulation and the etiology of many diseases. Statistical methods, particularly Bayesian networks, have been widely used to model cell signaling, mostly for model organisms and with focus on uncovering connectivity rather than inferring aberrations. Extensions to mammalian systems have not yielded compelling results, due likely to greatly increased complexity and limited proteomic measurements in vivo. In this study, we propose a comprehensive statistical model that is anchored to a predefined core topology, has a limited complexity due to parameter sharing and uses microarray data of mRNA transcripts as the only observable components of signaling. Specifically, we account for cell heterogeneity and a multilevel process, representing signaling as a Bayesian network at the cell level, modeling measurements as ensemble averages at the tissue level, and incorporating patient-to-patient differences at the population level. Motivated by the goal of identifying individual protein abnormalities as potential therapeutical targets, we applied our method to the RAS-RAF network using a breast cancer study with 118 patients. We demonstrated rigorous statistical inference, established reproducibility through simulations and the ability to recover receptor status from available microarray data.

Detection of treatment-induced changes in signaling pathways in gastrointestinal stromal tumors using transcriptomic data

Cell signaling plays a central role in the etiology of cancer. Numerous therapeutics in use or under development target signaling proteins; however, off-target effects often limit assignment of positive clinical response to the intended target. As direct measurements of signaling protein activity are not generally feasible during treatment, there is a need for more powerful methods to determine if therapeutics inhibit their targets and when off-target effects occur. We have used the Bayesian Decomposition algorithm and data on transcriptional regulation to create a novel methodology, Differential Expression for Signaling Determination (DESIDE), for inferring signaling activity from microarray measurements. We applied DESIDE to deduce signaling activity in gastrointestinal stromal tumor cell lines treated with the targeted therapeutic imatinib mesylate (Gleevec). We detected the expected reduced activity in the KIT pathway, as well as unexpected changes in the p53 pathway. Pursuing these findings, we have determined that imatinib-induced DNA damage is responsible for the increased activity of p53, identifying a novel off-target activity for this drug. We then used DESIDE on data from resected, post-imatinib treatment tumor samples and identified a pattern in these tumors similar to that at late time points in the cell lines, and this pattern correlated with initial clinical response. The pattern showed increased activity of ETS domain-containing protein Elk-1 and signal transducers and activators of transcription 3 transcription factors, which are associated with the growth of side population cells. DESIDE infers the global reprogramming of signaling networks during treatment, permitting treatment modification that leverages ongoing drug development efforts, which is crucial for personalized medicine.