Development and validation of a reliable DNA copy-number-based machine learning algorithm (CopyClust) for breast cancer integrative cluster classification.

The Integrative Cluster subtypes (IntClusts) provide a framework for the classification of breast cancer tumors into 10 distinct groups based on copy number and gene expression, each with unique biological drivers of disease and clinical prognoses. Gene expression data is often lacking, and accurate classification of samples into IntClusts with copy number data alone is essential. Current classification methods achieve low accuracy when gene expression data are absent, warranting the development of new approaches to IntClust classification. Copy number data from 1980 breast cancer samples from METABRIC was used to train multiclass XGBoost machine learning algorithms (CopyClust). A piecewise constant fit was applied to the average copy number profile of each IntClust and unique breakpoints across the 10 profiles were identified and converted into ~ 500 genomic regions used as features for CopyClust. These models consisted of two approaches: a 10-class model with the final IntClust label predicted by a single multiclass model and a 6-class model with binary reclassification in which four pairs of IntClusts were combined for initial multiclass classification. Performance was validated on the TCGA dataset, with copy number data generated from both SNP arrays and WES platforms. CopyClust achieved 81% and 79% overall accuracy with the TCGA SNP and WES datasets, respectively, a nine-percentage point or greater improvement in overall IntClust subtype classification accuracy. CopyClust achieves a significant improvement over current methods in classification accuracy of IntClust subtypes for samples without available gene expression data and is an easily implementable algorithm for IntClust classification of breast cancer samples with copy number data.

Human variation in population-wide gene expression data predicts gene perturbation phenotype.

Population-scale datasets of healthy individuals capture genetic and environmental factors influencing gene expression. The expression variance of a gene of interest (GOI) can be exploited to set up a quasi loss- or gain-of-function "in population" experiment. We describe here an approach, huva (human variation), taking advantage of population-scale multi-layered data to infer gene function and relationships between phenotypes and expression. Within a reference dataset, huva derives two experimental groups with LOW or HIGH expression of the GOI, enabling the subsequent comparison of their transcriptional profile and functional parameters. We demonstrate that this approach robustly identifies the phenotypic relevance of a GOI allowing the stratification of genes according to biological functions, and we generalize this concept to almost 16,000 genes in the human transcriptome. Additionally, we describe how huva predicts monocytes to be the major cell type in the pathophysiology of STAT1 mutations, evidence validated in a clinical cohort.

Decoding mechanism of action and sensitivity to drug candidates from integrated transcriptome and chromatin state.

Omics-based technologies are driving major advances in precision medicine, but efforts are still required to consolidate their use in drug discovery. In this work, we exemplify the use of multi-omics to support the development of 3-chloropiperidines, a new class of candidate anticancer agents. Combined analyses of transcriptome and chromatin accessibility elucidated the mechanisms underlying sensitivity to test agents. Furthermore, we implemented a new versatile strategy for the integration of RNA- and ATAC-seq (Assay for Transposase-Accessible Chromatin) data, able to accelerate and extend the standalone analyses of distinct omic layers. This platform guided the construction of a perturbation-informed basal signature predicting cancer cell lines' sensitivity and to further direct compound development against specific tumor types. Overall, this approach offers a scalable pipeline to support the early phases of drug discovery, understanding of mechanisms, and potentially inform the positioning of therapeutics in the clinic.

Tailored Bayes: a risk modeling framework under unequal misclassification costs.

Risk prediction models are a crucial tool in healthcare. Risk prediction models with a binary outcome (i.e., binary classification models) are often constructed using methodology which assumes the costs of different classification errors are equal. In many healthcare applications, this assumption is not valid, and the differences between misclassification costs can be quite large. For instance, in a diagnostic setting, the cost of misdiagnosing a person with a life-threatening disease as healthy may be larger than the cost of misdiagnosing a healthy person as a patient. In this article, we present Tailored Bayes (TB), a novel Bayesian inference framework which "tailors" model fitting to optimize predictive performance with respect to unbalanced misclassification costs. We use simulation studies to showcase when TB is expected to outperform standard Bayesian methods in the context of logistic regression. We then apply TB to three real-world applications, a cardiac surgery, a breast cancer prognostication task, and a breast cancer tumor classification task and demonstrate the improvement in predictive performance over standard methods.

High-dimensional regression in practice: an empirical study of finite-sample prediction, variable selection and ranking.

Penalized likelihood approaches are widely used for high-dimensional regression. Although many methods have been proposed and the associated theory is now well developed, the relative efficacy of different approaches in finite-sample settings, as encountered in practice, remains incompletely understood. There is therefore a need for empirical investigations in this area that can offer practical insight and guidance to users. In this paper, we present a large-scale comparison of penalized regression methods. We distinguish between three related goals: prediction, variable selection and variable ranking. Our results span more than 2300 data-generating scenarios, including both synthetic and semisynthetic data (real covariates and simulated responses), allowing us to systematically consider the influence of various factors (sample size, dimensionality, sparsity, signal strength and multicollinearity). We consider several widely used approaches (Lasso, Adaptive Lasso, Elastic Net, Ridge Regression, SCAD, the Dantzig Selector and Stability Selection). We find considerable variation in performance between methods. Our results support a "no panacea" view, with no unambiguous winner across all scenarios or goals, even in this restricted setting where all data align well with the assumptions underlying the methods. The study allows us to make some recommendations as to which approaches may be most (or least) suitable given the goal and some data characteristics. Our empirical results complement existing theory and provide a resource to compare methods across a range of scenarios and metrics.

Scalable Prediction of Acute Myeloid Leukemia Using High-Dimensional Machine Learning and Blood Transcriptomics.

Acute myeloid leukemia (AML) is a severe, mostly fatal hematopoietic malignancy. We were interested in whether transcriptomic-based machine learning could predict AML status without requiring expert input. Using 12,029 samples from 105 different studies, we present a large-scale study of machine learning-based prediction of AML in which we address key questions relating to the combination of machine learning and transcriptomics and their practical use. We find data-driven, high-dimensional approaches-in which multivariate signatures are learned directly from genome-wide data with no prior knowledge-to be accurate and robust. Importantly, these approaches are highly scalable with low marginal cost, essentially matching human expert annotation in a near-automated workflow. Our results support the notion that transcriptomics combined with machine learning could be used as part of an integrated -omics approach wherein risk prediction, differential diagnosis, and subclassification of AML are achieved by genomics while diagnosis could be assisted by transcriptomic-based machine learning.

The joint lasso: high-dimensional regression for group structured data.

We consider high-dimensional regression over subgroups of observations. Our work is motivated by biomedical problems, where subsets of samples, representing for example disease subtypes, may differ with respect to underlying regression models. In the high-dimensional setting, estimating a different model for each subgroup is challenging due to limited sample sizes. Focusing on the case in which subgroup-specific models may be expected to be similar but not necessarily identical, we treat subgroups as related problem instances and jointly estimate subgroup-specific regression coefficients. This is done in a penalized framework, combining an $\ell_1$ term with an additional term that penalizes differences between subgroup-specific coefficients. This gives solutions that are globally sparse but that allow information-sharing between the subgroups. We present algorithms for estimation and empirical results on simulated data and using Alzheimer's disease, amyotrophic lateral sclerosis, and cancer datasets. These examples demonstrate the gains joint estimation can offer in prediction as well as in providing subgroup-specific sparsity patterns.

Statistical Network Inference for Time-Varying Molecular Data with Dynamic Bayesian Networks.

In this chapter, we review the problem of network inference from time-course data, focusing on a class of graphical models known as dynamic Bayesian networks (DBNs). We discuss the relationship of DBNs to models based on ordinary differential equations, and consider extensions to nonlinear time dynamics. We provide an introduction to time-varying DBN models, which allow for changes to the network structure and parameters over time. We also discuss causal perspectives on network inference, including issues around model semantics that can arise due to missing variables. We present a case study of applying time-varying DBNs to gene expression measurements over the life cycle of Drosophila melanogaster. We finish with a discussion of future perspectives, including possible applications of time-varying network inference to single-cell gene expression data.

Causal Learning via Manifold Regularization.

This paper frames causal structure estimation as a machine learning task. The idea is to treat indicators of causal relationships between variables as 'labels' and to exploit available data on the variables of interest to provide features for the labelling task. Background scientific knowledge or any available interventional data provide labels on some causal relationships and the remainder are treated as unlabelled. To illustrate the key ideas, we develop a distance-based approach (based on bivariate histograms) within a manifold regularization framework. We present empirical results on three different biological data sets (including examples where causal effects can be verified by experimental intervention), that together demonstrate the efficacy and general nature of the approach as well as its simplicity from a user's point of view.

Correction: Decoupling of the PI3K Pathway via Mutation Necessitates Combinatorial Treatment in HER2+ Breast Cancer.

[This corrects the article DOI: 10.1371/journal.pone.0133219.].

Molecular heterogeneity at the network level: high-dimensional testing, clustering and a TCGA case study.

MOTIVATION: Molecular pathways and networks play a key role in basic and disease biology. An emerging notion is that networks encoding patterns of molecular interplay may themselves differ between contexts, such as cell type, tissue or disease (sub)type. However, while statistical testing of differences in mean expression levels has been extensively studied, testing of network differences remains challenging. Furthermore, since network differences could provide important and biologically interpretable information to identify molecular subgroups, there is a need to consider the unsupervised task of learning subgroups and networks that define them. This is a nontrivial clustering problem, with neither subgroups nor subgroup-specific networks known at the outset. RESULTS: We leverage recent ideas from high-dimensional statistics for testing and clustering in the network biology setting. The methods we describe can be applied directly to most continuous molecular measurements and networks do not need to be specified beforehand. We illustrate the ideas and methods in a case study using protein data from The Cancer Genome Atlas (TCGA). This provides evidence that patterns of interplay between signalling proteins differ significantly between cancer types. Furthermore, we show how the proposed approaches can be used to learn subtypes and the molecular networks that define them. AVAILABILITY AND IMPLEMENTATION: As the Bioconductor package nethet. CONTACT: staedler.n@gmail.com or sach.mukherjee@dzne.de. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Context Specificity in Causal Signaling Networks Revealed by Phosphoprotein Profiling.

Signaling networks downstream of receptor tyrosine kinases are among the most extensively studied biological networks, but new approaches are needed to elucidate causal relationships between network components and understand how such relationships are influenced by biological context and disease. Here, we investigate the context specificity of signaling networks within a causal conceptual framework using reverse-phase protein array time-course assays and network analysis approaches. We focus on a well-defined set of signaling proteins profiled under inhibition with five kinase inhibitors in 32 contexts: four breast cancer cell lines (MCF7, UACC812, BT20, and BT549) under eight stimulus conditions. The data, spanning multiple pathways and comprising ∼70,000 phosphoprotein and ∼260,000 protein measurements, provide a wealth of testable, context-specific hypotheses, several of which we experimentally validate. Furthermore, the data provide a unique resource for computational methods development, permitting empirical assessment of causal network learning in a complex, mammalian setting.

Inferring causal molecular networks: empirical assessment through a community-based effort.

It remains unclear whether causal, rather than merely correlational, relationships in molecular networks can be inferred in complex biological settings. Here we describe the HPN-DREAM network inference challenge, which focused on learning causal influences in signaling networks. We used phosphoprotein data from cancer cell lines as well as in silico data from a nonlinear dynamical model. Using the phosphoprotein data, we scored more than 2,000 networks submitted by challenge participants. The networks spanned 32 biological contexts and were scored in terms of causal validity with respect to unseen interventional data. A number of approaches were effective, and incorporating known biology was generally advantageous. Additional sub-challenges considered time-course prediction and visualization. Our results suggest that learning causal relationships may be feasible in complex settings such as disease states. Furthermore, our scoring approach provides a practical way to empirically assess inferred molecular networks in a causal sense.

A Gibbs Sampler for Learning DAGs.

We propose a Gibbs sampler for structure learning in directed acyclic graph (DAG) models. The standard Markov chain Monte Carlo algorithms used for learning DAGs are random-walk Metropolis-Hastings samplers. These samplers are guaranteed to converge asymptotically but often mix slowly when exploring the large graph spaces that arise in structure learning. In each step, the sampler we propose draws entire sets of parents for multiple nodes from the appropriate conditional distribution. This provides an efficient way to make large moves in graph space, permitting faster mixing whilst retaining asymptotic guarantees of convergence. The conditional distribution is related to variable selection with candidate parents playing the role of covariates or inputs. We empirically examine the performance of the sampler using several simulated and real data examples. The proposed method gives robust results in diverse settings, outperforming several existing Bayesian and frequentist methods. In addition, our empirical results shed some light on the relative merits of Bayesian and constraint-based methods for structure learning.

Decoupling of the PI3K Pathway via Mutation Necessitates Combinatorial Treatment in HER2+ Breast Cancer.

We report here on experimental and theoretical efforts to determine how best to combine drugs that inhibit HER2 and AKT in HER2(+) breast cancers. We accomplished this by measuring cellular and molecular responses to lapatinib and the AKT inhibitors (AKTi) GSK690693 and GSK2141795 in a panel of 22 HER2(+) breast cancer cell lines carrying wild type or mutant PIK3CA. We observed that combinations of lapatinib plus AKTi were synergistic in HER2(+)/PIK3CA(mut) cell lines but not in HER2(+)/PIK3CA(wt) cell lines. We measured changes in phospho-protein levels in 15 cell lines after treatment with lapatinib, AKTi or lapatinib + AKTi to shed light on the underlying signaling dynamics. This revealed that p-S6RP levels were less well attenuated by lapatinib in HER2(+)/PIK3CA(mut) cells compared to HER2(+)/PIK3CAwt cells and that lapatinib + AKTi reduced p-S6RP levels to those achieved in HER2(+)/PIK3CA(wt) cells with lapatinib alone. We also found that that compensatory up-regulation of p-HER3 and p-HER2 is blunted in PIK3CA(mut) cells following lapatinib + AKTi treatment. Responses of HER2(+) SKBR3 cells transfected with lentiviruses carrying control or PIK3CA(mut )sequences were similar to those observed in HER2(+)/PIK3CA(mut) cell lines but not in HER2(+)/PIK3CA(wt) cell lines. We used a nonlinear ordinary differential equation model to support the idea that PIK3CA mutations act as downstream activators of AKT that blunt lapatinib inhibition of downstream AKT signaling and that the effects of PIK3CA mutations can be countered by combining lapatinib with an AKTi. This combination does not confer substantial benefit beyond lapatinib in HER2+/PIK3CA(wt) cells.

Genome-wide perturbations by miRNAs map onto functional cellular pathways, identifying regulators of chromatin modifiers.

BACKGROUND: Regulation of gene expression by microRNAs (miRNAs) is critical for determining cellular fate and function. Dysregulation of miRNA expression contributes to the development and progression of multiple diseases. miRNA can target multiple mRNAs, making deconvolution of the effects of miRNA challenging and the complexity of regulation of cellular pathways by miRNAs at the functional protein level remains to be elucidated. Therefore, we sought to determine the effects of expression of miRNAs in breast and ovarian cancer cells on cellular pathways by measuring systems-wide miRNA perturbations to protein and phosphoproteins. METHODS: We measure protein level changes by reverse-phase protein array (RPPA) in MDA-MB-231, SKOV3.ip1 and HEYA8 cancer cell lines transfected by a library of 879 human miRNA mimics. RESULTS: The effects of multiple miRNAs-protein networks converged in five broad functional clusters of miRNA, suggesting a broad overlap of miRNA action on cellular pathways. Detailed analysis of miRNA clusters revealed novel miRNA/cell cycle protein networks, which we functionally validated. De novo phosphoprotein network estimation using Gaussian graphical modeling, using no priors, revealed known and novel protein interplay, which we also observed in patient ovarian tumor proteomic data. We identified several miRNAs that have pluripotent activities across multiple cellular pathways. In particular we studied miR-365a whose expression is associated with poor survival across several cancer types and demonstrated that anti-miR-365 significantly reduced tumor formation in animal models. CONCLUSIONS: Mapping of miRNA-induced protein and phosphoprotein changes onto pathways revealed new miRNA-cellular pathway connectivity, paving the way for targeting of dysregulated pathways with potential miRNA-based therapeutics.

Multivariate gene-set testing based on graphical models.

The identification of predefined groups of genes ("gene-sets") which are differentially expressed between two conditions ("gene-set analysis", or GSA) is a very popular analysis in bioinformatics. GSA incorporates biological knowledge by aggregating over genes that are believed to be functionally related. This can enhance statistical power over analyses that consider only one gene at a time. However, currently available GSA approaches are based on univariate two-sample comparison of single genes. This means that they cannot test for multivariate hypotheses such as differences in covariance structure between the two conditions. Yet interplay between genes is a central aspect of biological investigation and it is likely that such interplay may differ between conditions. This paper proposes a novel approach for gene-set analysis that allows for truly multivariate hypotheses, in particular differences in gene-gene networks between conditions. Testing hypotheses concerning networks is challenging due the nature of the underlying estimation problem. Our starting point is a recent, general approach for high-dimensional two-sample testing. We refine the approach and show how it can be used to perform multivariate, network-based gene-set testing. We validate the approach in simulated examples and show results using high-throughput data from several studies in cancer biology.

Causal network inference using biochemical kinetics.

MOTIVATION: Networks are widely used as structural summaries of biochemical systems. Statistical estimation of networks is usually based on linear or discrete models. However, the dynamics of biochemical systems are generally non-linear, suggesting that suitable non-linear formulations may offer gains with respect to causal network inference and aid in associated prediction problems. RESULTS: We present a general framework for network inference and dynamical prediction using time course data that is rooted in non-linear biochemical kinetics. This is achieved by considering a dynamical system based on a chemical reaction graph with associated kinetic parameters. Both the graph and kinetic parameters are treated as unknown; inference is carried out within a Bayesian framework. This allows prediction of dynamical behavior even when the underlying reaction graph itself is unknown or uncertain. Results, based on (i) data simulated from a mechanistic model of mitogen-activated protein kinase signaling and (ii) phosphoproteomic data from cancer cell lines, demonstrate that non-linear formulations can yield gains in causal network inference and permit dynamical prediction and uncertainty quantification in the challenging setting where the reaction graph is unknown. AVAILABILITY AND IMPLEMENTATION: MATLAB R2014a software is available to download from warwick.ac.uk/chrisoates. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

A pan-cancer proteomic perspective on The Cancer Genome Atlas.

Protein levels and function are poorly predicted by genomic and transcriptomic analysis of patient tumours. Therefore, direct study of the functional proteome has the potential to provide a wealth of information that complements and extends genomic, epigenomic and transcriptomic analysis in The Cancer Genome Atlas (TCGA) projects. Here we use reverse-phase protein arrays to analyse 3,467 patient samples from 11 TCGA 'Pan-Cancer' diseases, using 181 high-quality antibodies that target 128 total proteins and 53 post-translationally modified proteins. The resultant proteomic data are integrated with genomic and transcriptomic analyses of the same samples to identify commonalities, differences, emergent pathways and network biology within and across tumour lineages. In addition, tissue-specific signals are reduced computationally to enhance biomarker and target discovery spanning multiple tumour lineages. This integrative analysis, with an emphasis on pathways and potentially actionable proteins, provides a framework for determining the prognostic, predictive and therapeutic relevance of the functional proteome.