kMEn: Analyzing noisy and bidirectional transcriptional pathway responses in single subjects

Accounting for extra-binomial variability with differentially expressed genetic pathway data: a collaborative bioinformatic study

We describe a collaborative project involving faculty and students in a university bioinformatics/biostatistics center. The project focuses on identification of differentially expressed gene sets ("pathways") in subjects expressing a disease state, medical intervention, or other distinguishable condition. The key feature of the endeavor is the data structure presented to the team: a single cohort of subjects with two samples taken from each subject - one for each of two differing conditions without replication. This particular structure leads to essentially a cohort of 2 × 2 contingency tables, where each table compares the differential gene state with the pathway condition. Recognizing that correlations both within and between pathway responses can disrupt standard 2 × 2 table analytics, we develop methods for analyzing this data structure in the presence of complicated intra-table correlations. These provide some convenient approaches for this problem, using design effect adjustments from sample survey theory and manipulations of the summary 2 × 2 table counts. Monte Carlo simulations show that the methods operate extremely well, validating their use in practice. In the end, the collaborative connections among the team members led to solutions no one of us would have envisioned separately.

'Single-subject studies'-derived analyses unveil altered biomechanisms between very small cohorts: implications for rare diseases

MOTIVATION

Identifying altered transcripts between very small human cohorts is particularly challenging and is compounded by the low accrual rate of human subjects in rare diseases or sub-stratified common disorders. Yet, single-subject studies (S3) can compare paired transcriptome samples drawn from the same patient under two conditions (e.g. treated versus pre-treatment) and suggest patient-specific responsive biomechanisms based on the overrepresentation of functionally defined gene sets. These improve statistical power by: (i) reducing the total features tested and (ii) relaxing the requirement of within-cohort uniformity at the transcript level. We propose Inter-N-of-1, a novel method, to identify meaningful differences between very small cohorts by using the effect size of 'single-subject-study'-derived responsive biological mechanisms.

RESULTS

In each subject, Inter-N-of-1 requires applying previously published S3-type N-of-1-pathways MixEnrich to two paired samples (e.g. diseased versus unaffected tissues) for determining patient-specific enriched genes sets: Odds Ratios (S3-OR) and S3-variance using Gene Ontology Biological Processes. To evaluate small cohorts, we calculated the precision and recall of Inter-N-of-1 and that of a control method (GLM+EGS) when comparing two cohorts of decreasing sizes (from 20 versus 20 to 2 versus 2) in a comprehensive six-parameter simulation and in a proof-of-concept clinical dataset. In simulations, the Inter-N-of-1 median precision and recall are > 90% and >75% in cohorts of 3 versus 3 distinct subjects (regardless of the parameter values), whereas conventional methods outperform Inter-N-of-1 at sample sizes 9 versus 9 and larger. Similar results were obtained in the clinical proof-of-concept dataset.

AVAILABILITY AND IMPLEMENTATION

R software is available at Lussierlab.net/BSSD.

Interpretation of 'Omics dynamics in a single subject using local estimates of dispersion between two transcriptomes

Calculating Differentially Expressed Genes (DEGs) from RNA-sequencing requires replicates to estimate gene-wise variability, a requirement that is at times financially or physiologically infeasible in clinics. By imposing restrictive transcriptome-wide assumptions limiting inferential opportunities of conventional methods (edgeR, NOISeq-sim, DESeq, DEGseq), comparing two conditions without replicates (TCWR) has been proposed, but not evaluated. Under TCWR conditions (e.g., unaffected tissue vs. tumor), differences of transformed expression of the proposed individualized DEG (iDEG) method follow a distribution calculated across a local partition of related transcripts at baseline expression; thereafter the probability of each DEG is estimated by empirical Bayes with local false discovery rate control using a two-group mixture model. In extensive simulation studies of TCWR methods, iDEG and NOISeq are more accurate at 5%<DEGs<20% (precision>90%, recall>75%, false_positive_rate<1%) and 30%<DEGs<40% (precision=recall~90%), respectively. The proposed iDEG method borrows localized distribution information from the same individual, a strategy that improves accuracy to compare transcriptomes in absence of replicates at low DEGsconditions. http://www.lussiergroup.org/publications/iDEG.

Evaluating single-subject study methods for personal transcriptomic interpretations to advance precision medicine

Background

Gene expression profiling has benefited medicine by providing clinically relevant insights at the molecular candidate and systems levels. However, to adopt a more ‘precision’ approach that integrates individual variability including ‘omics data into risk assessments, diagnoses, and therapeutic decision making, whole transcriptome expression needs to be interpreted meaningfully for single subjects. We propose an “all-against-one” framework that uses biological replicates in isogenic conditions for testing differentially expressed genes (DEGs) in a single subject (ss) in the absence of an appropriate external reference standard or replicates. To evaluate our proposed “all-against-one” framework, we construct reference standards (RSs) with five conventional replicate-anchored analyses (NOISeq, DEGseq, edgeR, DESeq, DESeq2) and the remainder were treated separately as single-subject sample pairs for ss analyses (without replicates).

Results

Eight ss methods (NOISeq, DEGseq, edgeR, mixture model, DESeq, DESeq2, iDEG, and ensemble) for identifying genes with differential expression were compared in Yeast (parental line versus snf2 deletion mutant; n = 42/condition) and a MCF7 breast-cancer cell line (baseline versus stimulated with estradiol; n = 7/condition). Receiver-operator characteristic (ROC) and precision-recall plots were determined for eight ss methods against each of the five RSs in both datasets. Consistent with prior analyses of these data, ~ 50% and ~ 15% DEGs were obtained in Yeast and MCF7 datasets respectively, regardless of the RSs method. NOISeq, edgeR, and DESeq were the most concordant for creating a RS. Single-subject versions of NOISeq, DEGseq, and an ensemble learner achieved the best median ROC-area-under-the-curve to compare two transcriptomes without replicates regardless of the RS method and dataset (> 90% in Yeast, > 0.75 in MCF7). Further, distinct specific single-subject methods perform better according to different proportions of DEGs.

Conclusions

The “all-against-one” framework provides a honest evaluation framework for single-subject DEG studies since these methods are evaluated, by design, against reference standards produced by unrelated DEG methods. The ss-ensemble method was the only one to reliably produce higher accuracies in all conditions tested in this conservative evaluation framework. However, single-subject methods for identifying DEGs from paired samples need improvement, as no method performed with precision> 90% and obtained moderate levels of recall.

http://www.lussiergroup.org/publications/EnsembleBiomarker

Developing a 'personalome' for precision medicine: emerging methods that compute interpretable effect sizes from single-subject transcriptomes

The development of computational methods capable of analyzing -omics data at the individual level is critical for the success of precision medicine. Although unprecedented opportunities now exist to gather data on an individual's -omics profile ('personalome'), interpreting and extracting meaningful information from single-subject -omics remain underdeveloped, particularly for quantitative non-sequence measurements, including complete transcriptome or proteome expression and metabolite abundance. Conventional bioinformatics approaches have largely been designed for making population-level inferences about 'average' disease processes; thus, they may not adequately capture and describe individual variability. Novel approaches intended to exploit a variety of -omics data are required for identifying individualized signals for meaningful interpretation. In this review-intended for biomedical researchers, computational biologists and bioinformaticians-we survey emerging computational and translational informatics methods capable of constructing a single subject's 'personalome' for predicting clinical outcomes or therapeutic responses, with an emphasis on methods that provide interpretable readouts. Key points: (i) the single-subject analytics of the transcriptome shows the greatest development to date and, (ii) the methods were all validated in simulations, cross-validations or independent retrospective data sets. This survey uncovers a growing field that offers numerous opportunities for the development of novel validation methods and opens the door for future studies focusing on the interpretation of comprehensive 'personalomes' through the integration of multiple -omics, providing valuable insights into individual patient outcomes and treatments.

A Single-Subject Method to Detect Pathways Enriched With Alternatively Spliced Genes

RNA-Sequencing data offers an opportunity to enable precision medicine, but most methods rely on gene expression alone. To date, no methodology exists to identify and interpret alternative splicing patterns within pathways for an individual patient. This study develops methodology and conducts computational experiments to test the hypothesis that pathway aggregation of subject-specific alternatively spliced genes (ASGs) can inform upon disease mechanisms and predict survival. We propose the N-of-1-pathways Alternatively Spliced (N1PAS) method that takes an individual patient’s paired-sample RNA-Seq isoform expression data (e.g., tumor vs. non-tumor, before-treatment vs. during-therapy) and pathway annotations as inputs. N1PAS quantifies the degree of alternative splicing via Hellinger distances followed by two-stage clustering to determine pathway enrichment. We provide a clinically relevant “odds ratio” along with statistical significance to quantify pathway enrichment. We validate our method in clinical samples and find that our method selects relevant pathways (p < 0.05 in 4/6 data sets). Extensive Monte Carlo studies show N1PAS powerfully detects pathway enrichment of ASGs while adequately controlling false discovery rates. Importantly, our studies also unveil highly heterogeneous single-subject alternative splicing patterns that cohort-based approaches overlook. Finally, we apply our patient-specific results to predict cancer survival (FDR < 20%) while providing diagnostics in pursuit of translating transcriptome data into clinically actionable information. Software available at https://github.com/grizant/n1pas/tree/master.

The Emerging Potential for Network Analysis to Inform Precision Cancer Medicine

Precision cancer medicine promises to tailor clinical decisions to patients using genomic information. Indeed, successes of drugs targeting genetic alterations in tumors, such as imatinib that targets BCR-ABL in chronic myelogenous leukemia, have demonstrated the power of this approach. However, biological systems are complex, and patients may differ not only by the specific genetic alterations in their tumor, but also by more subtle interactions among such alterations. Systems biology and more specifically, network analysis, provides a framework for advancing precision medicine beyond clinical actionability of individual mutations. Here we discuss applications of network analysis to study tumor biology, early methods for N-of-1 tumor genome analysis, and the path for such tools to the clinic.

Patient similarity for precision medicine: A systematic review

Evidence-based medicine is the most prevalent paradigm adopted by physicians. Clinical practice guidelines typically define a set of recommendations together with eligibility criteria that restrict their applicability to a specific group of patients. The ever-growing size and availability of health-related data is currently challenging the broad definitions of guideline-defined patient groups. Precision medicine leverages on genetic, phenotypic, or psychosocial characteristics to provide precise identification of patient subsets for treatment targeting. Defining a patient similarity measure is thus an essential step to allow stratification of patients into clinically-meaningful subgroups. The present review investigates the use of patient similarity as a tool to enable precision medicine. 279 articles were analyzed along four dimensions: data types considered, clinical domains of application, data analysis methods, and translational stage of findings. Cancer-related research employing molecular profiling and standard data analysis techniques such as clustering constitute the majority of the retrieved studies. Chronic and psychiatric diseases follow as the second most represented clinical domains. Interestingly, almost one quarter of the studies analyzed presented a novel methodology, with the most advanced employing data integration strategies and being portable to different clinical domains. Integration of such techniques into decision support systems constitutes and interesting trend for future research.

Emergence of pathway-level composite biomarkers from converging gene set signals of heterogeneous transcriptomic responses

Recent precision medicine initiatives have led to the expectation of improved clinical decisionmaking anchored in genomic data science. However, over the last decade, only a handful of new single-gene product biomarkers have been translated to clinical practice (FDA approved) in spite of considerable discovery efforts deployed and a plethora of transcriptomes available in the Gene Expression Omnibus. With this modest outcome of current approaches in mind, we developed a pilot simulation study to demonstrate the untapped benefits of developing disease detection methods for cases where the true signal lies at the pathway level, even if the pathway's gene expression alterations may be heterogeneous across patients. In other words, we relaxed the crosspatient homogeneity assumption from the transcript level (cohort assumptions of deregulated gene expression) to the pathway level (assumptions of deregulated pathway expression). Furthermore, we have expanded previous single-subject (SS) methods into cohort analyses to illustrate the benefit of accounting for an individual's variability in cohort scenarios. We compare SS and cohort-based (CB) techniques under 54 distinct scenarios, each with 1,000 simulations, to demonstrate that the emergence of a pathway-level signal occurs through the summative effect of its altered gene expression, heterogeneous across patients. Studied variables include pathway gene set size, fraction of expressed gene responsive within gene set, fraction of expressed gene responsive up- vs down-regulated, and cohort size. We demonstrated that our SS approach was uniquely suited to detect signals in heterogeneous populations in which individuals have varying levels of baseline risks that are simultaneously confounded by patient-specific "genome -by-environment" interactions (G×E). Area under the precision-recall curve of the SS approach far surpassed that of the CB (1st quartile, median, 3rd quartile: SS = 0.94, 0.96, 0.99; CB= 0.50, 0.52, 0.65). We conclude that single-subject pathway detection methods are uniquely suited for consistently detecting pathway dysregulation by the inclusion of a patient's individual variability. http://www.lussiergroup.org/publications/PathwayMarker/.

N-of-1-pathways MixEnrich: advancing precision medicine via single-subject analysis in discovering dynamic changes of transcriptomes

Background

Transcriptome analytic tools are commonly used across patient cohorts to develop drugs and predict clinical outcomes. However, as precision medicine pursues more accurate and individualized treatment decisions, these methods are not designed to address single-patient transcriptome analyses. We previously developed and validated the N-of-1-pathways framework using two methods, Wilcoxon and Mahalanobis Distance (MD), for personal transcriptome analysis derived from a pair of samples of a single patient. Although, both methods uncover concordantly dysregulated pathways, they are not designed to detect dysregulated pathways with up- and down-regulated genes (bidirectional dysregulation) that are ubiquitous in biological systems.

Results

We developed N-of-1-pathways MixEnrich, a mixture model followed by a gene set enrichment test, to uncover bidirectional and concordantly dysregulated pathways one patient at a time. We assess its accuracy in a comprehensive simulation study and in a RNA-Seq data analysis of head and neck squamous cell carcinomas (HNSCCs). In presence of bidirectionally dysregulated genes in the pathway or in presence of high background noise, MixEnrich substantially outperforms previous single-subject transcriptome analysis methods, both in the simulation study and the HNSCCs data analysis (ROC Curves; higher true positive rates; lower false positive rates). Bidirectional and concordant dysregulated pathways uncovered by MixEnrich in each patient largely overlapped with the quasi-gold standard compared to other single-subject and cohort-based transcriptome analyses.

Conclusion

The greater performance of MixEnrich presents an advantage over previous methods to meet the promise of providing accurate personal transcriptome analysis to support precision medicine at point of care.

Electronic supplementary material

The online version of this article (doi:10.1186/s12920-017-0263-4) contains supplementary material, which is available to authorized users.