Model-agnostic explanations for survival prediction models.

Advanced machine learning methods capable of capturing complex and nonlinear relationships can be used in biomedical research to accurately predict time-to-event outcomes. However, these methods have been criticized as "black boxes" that are not interpretable and thus are difficult to trust in making important clinical decisions. Explainable machine learning proposes the use of model-agnostic explainers that can be applied to predictions from any complex model. These explainers describe how a patient's characteristics are contributing to their prediction, and thus provide insight into how the model is arriving at that prediction. The specific application of these explainers to survival prediction models can be used to obtain explanations for (i) survival predictions at particular follow-up times, and (ii) a patient's overall predicted survival curve. Here, we present a model-agnostic approach for obtaining these explanations from any survival prediction model. We extend the local interpretable model-agnostic explainer framework for classification outcomes to survival prediction models. Using simulated data, we assess the performance of the proposed approaches under various settings. We illustrate application of the new methodology using prostate cancer data.

Uncertainty-Aware Deep Learning Classification of Adamantinomatous Craniopharyngioma from Preoperative MRI.

Diagnosis of adamantinomatous craniopharyngioma (ACP) is predominantly determined through invasive pathological examination of a neurosurgical biopsy specimen. Clinical experts can distinguish ACP from Magnetic Resonance Imaging (MRI) with an accuracy of 86%, and 9% of ACP cases are diagnosed this way. Classification using deep learning (DL) provides a solution to support a non-invasive diagnosis of ACP through neuroimaging, but it is still limited in implementation, a major reason being the lack of predictive uncertainty representation. We trained and tested a DL classifier on preoperative MRI from 86 suprasellar tumor patients across multiple institutions. We then applied a Bayesian DL approach to calibrate our previously published ACP classifier, extending beyond point-estimate predictions to predictive distributions. Our original classifier outperforms random forest and XGBoost models in classifying ACP. The calibrated classifier underperformed our previously published results, indicating that the original model was overfit. Mean values of the predictive distributions were not informative regarding model uncertainty. However, the variance of predictive distributions was indicative of predictive uncertainty. We developed an algorithm to incorporate predicted values and the associated uncertainty to create a classification abstention mechanism. Our model accuracy improved from 80.8% to 95.5%, with a 34.2% abstention rate. We demonstrated that calibration of DL models can be used to estimate predictive uncertainty, which may enable clinical translation of artificial intelligence to support non-invasive diagnosis of brain tumors in the future.

EASL: A Framework for Designing, Implementing, and Evaluating ML Solutions in Clinical Healthcare Settings.

We introduce the Explainable Analytical Systems Lab (EASL) framework, an end-to-end solution designed to facilitate the development, implementation, and evaluation of clinical machine learning (ML) tools. EASL is highly versatile and applicable to a variety of contexts and includes resources for data management, ML model development, visualization and user interface development, service hosting, and usage analytics. To demonstrate its practical applications, we present the EASL framework in the context of a case study: designing and evaluating a deep learning classifier to predict diagnoses from medical imaging. The framework is composed of three modules, each with their own set of resources. The Workbench module stores data and develops initial ML models, the Canvas module contains a medical imaging viewer and web development framework, and the Studio module hosts the ML model and provides web analytics and support for conducting user studies. EASL encourages model developers to take a holistic view by integrating the model development, implementation, and evaluation into one framework, and thus ensures that models are both effective and reliable when used in a clinical setting. EASL contributes to our understanding of machine learning applied to healthcare by providing a comprehensive framework that makes it easier to develop and evaluate ML tools within a clinical setting.

The Iterative Design Process of an Explainable AI Application for Non-Invasive Diagnosis of CNS Tumors: A User-Centered Approach.

Artificial Intelligence (AI) is well-suited to help support complex decision-making tasks within clinical medicine, including clinical imaging applications like radiographic differential diagnosis of central nervous system (CNS) tumors. So far, there have been numerous examples of theoretical AI solutions for this space, for example, large-scale corporate efforts like IBM's Watson AI. However, clinical implementation remains limited due to factors related to the alignment of this technology in the clinical setting. User-Centered Design (UCD) is a design philosophy that focuses on developing tailored solutions for specific users or user groups. In this study, we applied UCD to develop an explainable AI tool to support clinicians in our use case. Through four design iterations, starting from basic functionality and visualizations, we progressed to functional prototypes in a realistic testing environment. We discuss our motivation and approach for each iteration, along with key insights gained. This UCD process has advanced our conceptual idea from feasibility testing to interactive functional AI interfaces designed for specific clinical and cognitive tasks. It has also provided us with directions to develop further an AI system for the non-invasive diagnosis of CNS tumors.

A Pipeline for the Implementation and Visualization of Explainable Machine Learning for Medical Imaging Using Radiomics Features.

Machine learning (ML) models have been shown to predict the presence of clinical factors from medical imaging with remarkable accuracy. However, these complex models can be difficult to interpret and are often criticized as "black boxes". Prediction models that provide no insight into how their predictions are obtained are difficult to trust for making important clinical decisions, such as medical diagnoses or treatment. Explainable machine learning (XML) methods, such as Shapley values, have made it possible to explain the behavior of ML algorithms and to identify which predictors contribute most to a prediction. Incorporating XML methods into medical software tools has the potential to increase trust in ML-powered predictions and aid physicians in making medical decisions. Specifically, in the field of medical imaging analysis the most used methods for explaining deep learning-based model predictions are saliency maps that highlight important areas of an image. However, they do not provide a straightforward interpretation of which qualities of an image area are important. Here, we describe a novel pipeline for XML imaging that uses radiomics data and Shapley values as tools to explain outcome predictions from complex prediction models built with medical imaging with well-defined predictors. We present a visualization of XML imaging results in a clinician-focused dashboard that can be generalized to various settings. We demonstrate the use of this workflow for developing and explaining a prediction model using MRI data from glioma patients to predict a genetic mutation.

Restoring invisible and abandoned trials of gabapentin for neuropathic pain: a clinical and methodological investigation.

INTRODUCTION: Gabapentin (Neurontin) is prescribed widely for conditions for which it has not been approved by regulators, including certain neuropathic pain conditions. There is limited evidence that gabapentin is safe and effective for the treatment of neuropathic pain. Published trial reports, and systematic reviews based on published trial reports, mislead patients and providers because information about gabapentin's harms has been published only partly. We confirmed that trials conducted by the drug developer have been abandoned, and we plan to conduct a restoration with support from the Restoring Invisible and Abandoned Trials Support Centre (https://restoringtrials.org/). METHODS AND ANALYSIS: In this study, we will analyse and report the harms that were observed in six trials of gabapentin, which have not been reported publicly (eg, in journal articles). We will use clinical study reports and individual participant data to identify and report the harms observed in each individual trial and to summarise the harms observed across all six trials. We will report all adverse events observed in the included trials by sharing deidentified data and summary tables on the Open Science Framework (https://osf.io/w8puv/). Additionally, we will produce a summary report that describes differences between the randomised groups in each trial and across trials for prespecified harms outcomes. ETHICS AND DISSEMINATION: We will use secondary data. This study was determined to be exempt from Institutional Review Board (IRB) review (protocol #1910607198).

Multiomic Predictors of Short-Term Weight Loss and Clinical Outcomes During a Behavioral-Based Weight Loss Intervention.

OBJECTIVE: Identifying predictors of weight loss and clinical outcomes may increase understanding of individual variability in weight loss response. We hypothesized that baseline multiomic features, including DNA methylation (DNAme), metabolomics, and gut microbiome, would be predictive of short-term changes in body weight and other clinical outcomes within a comprehensive weight loss intervention. METHODS: Healthy adults with overweight or obesity (n = 62, age 18-55 years, BMI 27-45 kg/m2 , 75.8% female) participated in a 1-year behavioral weight loss intervention. To identify baseline omic predictors of changes in clinical outcomes at 3 and 6 months, whole-blood DNAme, plasma metabolites, and gut microbial genera were analyzed. RESULTS: A network of multiomic relationships informed predictive models for 10 clinical outcomes (body weight, waist circumference, fat mass, hemoglobin A1c , homeostatic model assessment of insulin resistance, total cholesterol, triglycerides, C-reactive protein, leptin, and ghrelin) that changed significantly (P < 0.05). For eight of these, adjusted R2 ranged from 0.34 to 0.78. Our models identified specific DNAme sites, gut microbes, and metabolites that were predictive of variability in weight loss, waist circumference, and circulating triglycerides and that are biologically relevant to obesity and metabolic pathways. CONCLUSIONS: These data support the feasibility of using baseline multiomic features to provide insight for precision nutrition-based weight loss interventions.

Semantic Changepoint Detection for Finding Potentially Novel Research Publications.

How has the focus of research papers on a given disease changed over time? Identifying the papers at the cusps of change can help highlight the emergence of a new topic or a change in the direction of research. We present a generally applicable unsupervised approach to this question based on semantic changepoints within a given collection of research papers. We illustrate the approach by a range of examples based on a nascent corpus of literature on COVID-19 as well as subsets of papers from PubMed on the World Health Organization list of neglected tropical diseases. The software is freely available at: https://github.com/pdddinakar/SemanticChangepointDetection.

CANTARE: finding and visualizing network-based multi-omic predictive models.

BACKGROUND: One goal of multi-omic studies is to identify interpretable predictive models for outcomes of interest, with analytes drawn from multiple omes. Such findings could support refined biological insight and hypothesis generation. However, standard analytical approaches are not designed to be "ome aware." Thus, some researchers analyze data from one ome at a time, and then combine predictions across omes. Others resort to correlation studies, cataloging pairwise relationships, but lacking an obvious approach for cohesive and interpretable summaries of these catalogs. METHODS: We present a novel workflow for building predictive regression models from network neighborhoods in multi-omic networks. First, we generate pairwise regression models across all pairs of analytes from all omes, encoding the resulting "top table" of relationships in a network. Then, we build predictive logistic regression models using the analytes in network neighborhoods of interest. We call this method CANTARE (Consolidated Analysis of Network Topology And Regression Elements). RESULTS: We applied CANTARE to previously published data from healthy controls and patients with inflammatory bowel disease (IBD) consisting of three omes: gut microbiome, metabolomics, and microbial-derived enzymes. We identified 8 unique predictive models with AUC > 0.90. The number of predictors in these models ranged from 3 to 13. We compare the results of CANTARE to random forests and elastic-net penalized regressions, analyzing AUC, predictions, and predictors. CANTARE AUC values were competitive with those generated by random forests and  penalized regressions. The top 3 CANTARE models had a greater dynamic range of predicted probabilities than did random forests and penalized regressions (p-value = 1.35 × 10-5). CANTARE models were significantly more likely to prioritize predictors from multiple omes than were the alternatives (p-value = 0.005). We also showed that predictive models from a network based on pairwise models with an interaction term for IBD have higher AUC than predictive models built from a correlation network (p-value = 0.016). R scripts and a CANTARE User's Guide are available at https://sourceforge.net/projects/cytomelodics/files/CANTARE/ . CONCLUSION: CANTARE offers a flexible approach for building parsimonious, interpretable multi-omic models. These models yield quantitative and directional effect sizes for predictors and support the generation of hypotheses for follow-up investigation.

HIV Infection Is Associated with Loss of Anti-Inflammatory Alveolar Macrophages.

HIV type 1 is associated with pulmonary dysfunction that is exacerbated by cigarette smoke. Alveolar macrophages (AM) are the most prominent immune cell in the alveolar space. These cells play an important role in clearing inhaled pathogens and regulating the inflammatory environment; however, how HIV infection impacts AM phenotype and function is not well understood, in part because of their autofluorescence and the absence of well-defined surface markers. The main aim of this study was to evaluate the impact of HIV infection on human AM and to compare the effect of smoking on their phenotype and function. Time-of-flight mass cytometry and RNA sequencing were used to characterize macrophages from human bronchoalveolar lavage of HIV-infected and -uninfected smokers and nonsmokers. We found that the frequency of CD163+ anti-inflammatory AM was decreased, whereas CD163-CCR7+ proinflammatory AM were increased in HIV infection. HIV-mediated proinflammatory polarization was associated with increased levels of inflammatory cytokines and macrophage activation. Conversely, smoking heightened the inflammatory response evident by change in the expression of CXCR4 and TLR4. Altogether, these findings suggest that HIV infection, along with cigarette smoke, favors a proinflammatory macrophage phenotype associated with enhanced expression of inflammatory molecules. Further, this study highlights time-of-flight mass cytometry as a reliable method for immunophenotyping the highly autofluorescent cells present in the bronchoalveolar lavage of cigarette smokers.

ML-MEDIC: A Preliminary Study of an Interactive Visual Analysis Tool Facilitating Clinical Applications of Machine Learning for Precision Medicine.

Accessible interactive tools that integrate machine learning methods with clinical research and reduce the programming experience required are needed to move science forward. Here, we present Machine Learning for Medical Exploration and Data-Inspired Care (ML-MEDIC), a point-and-click, interactive tool with a visual interface for facilitating machine learning and statistical analyses in clinical research. We deployed ML-MEDIC in the American Heart Association (AHA) Precision Medicine Platform to provide secure internet access and facilitate collaboration. ML-MEDIC's efficacy for facilitating the adoption of machine learning was evaluated through two case studies in collaboration with clinical domain experts. A domain expert review was also conducted to obtain an impression of the usability and potential limitations.

VOLARE: visual analysis of disease-associated microbiome-immune system interplay.

BACKGROUND: Relationships between specific microbes and proper immune system development, composition, and function have been reported in a number of studies. However, researchers have discovered only a fraction of the likely relationships. "Omic" methodologies such as 16S ribosomal RNA (rRNA) sequencing and time-of-flight mass cytometry (CyTOF) immunophenotyping generate data that support generation of hypotheses, with the potential to identify additional relationships at a level of granularity ripe for further experimentation. Pairwise linear regressions between microbial and host immune features provide one approach for quantifying relationships between "omes", and the differences in these relationships across study cohorts or arms. This approach yields a top table of candidate results. However, the top table alone lacks the detail that domain experts such as microbiologists and immunologists need to vet candidate results for follow-up experiments. RESULTS: To support this vetting, we developed VOLARE (Visualization Of LineAr Regression Elements), a web application that integrates a searchable top table, small in-line graphs illustrating the fitted models, a network summarizing the top table, and on-demand detailed regression plots showing full sample-level detail. We applied VOLARE to three case studies-microbiome:cytokine data from fecal samples in human immunodeficiency virus (HIV), microbiome:cytokine data in inflammatory bowel disease and spondyloarthritis, and microbiome:immune cell data from gut biopsies in HIV. We present both patient-specific phenomena and relationships that differ by disease state. We also analyzed interaction data from system logs to characterize usage scenarios. This log analysis revealed that users frequently generated detailed regression plots, suggesting that this detail aids the vetting of results. CONCLUSIONS: Systematically integrating microbe:immune cell readouts through pairwise linear regressions and presenting the top table in an interactive environment supports the vetting of results for scientific relevance. VOLARE allows domain experts to control the analysis of their results, screening dozens of candidate relationships with ease. This interactive environment transcends the limitations of a static top table.

Functional intraepithelial lymphocyte changes in inflammatory bowel disease and spondyloarthritis have disease specific correlations with intestinal microbiota.

BACKGROUND: Dysbiosis occurs in spondyloarthritis (SpA) and inflammatory bowel disease (IBD), which is subdivided into Crohn's disease (CD) and ulcerative colitis (UC). The immunologic consequences of alterations in microbiota, however, have not been defined. Intraepithelial lymphocytes (IELs) are T cells within the intestinal epithelium that are in close contact with bacteria and are likely to be modulated by changes in microbiota. We examined differences in human gut-associated bacteria and tested correlation with functional changes in IELs in patients with axial SpA (axSpA), CD, or UC, and in controls. METHODS: We conducted a case-control study to evaluate IELs from pinch biopsies of grossly normal colonic tissue from subjects with biopsy-proven CD or UC, axSpA fulfilling Assessment of SpondyloArthritis International Society (ASAS) criteria and from controls during endoscopy. IELs were harvested and characterized by flow cytometry for cell surface markers. Secreted cytokines were measured by ELISA. Microbiome analysis was by 16S rRNA gene sequencing from rectal swabs. Statistical analyses were performed with the Kruskal-Wallis and Spearman's rank tests. RESULTS: The total number of IELs was significantly decreased in subjects with axSpA compared to those with IBD and controls, likely due to a decrease in TCRß+ IELs. We found strong, significant negative correlation between peripheral lymphocyte count and IEL number. IELs secreted significantly increased IL-1ß in patients with UC, significantly increased IL-17A and IFN-γ in patients with CD, and significantly increased TNF-α in patients with CD and axSpA as compared to other cohorts. For each disease subtype, IELs and IEL-produced cytokines were positively and negatively correlated with the relative abundance of multiple bacterial taxa. CONCLUSIONS: Our data indicate differences in IEL function among subjects with axSpA, CD, and UC compared to healthy controls. We propose that the observed correlation between altered microbiota and IEL function in these populations are relevant to the pathogenesis of axSpA and IBD, and discuss possible mechanisms. TRIAL REGISTRATION: ClinicalTrials.gov, NCT02389075 . Registered on 17 March 2015.

Phenotype-Specific Association of Single-Nucleotide Polymorphisms with Heart Failure and Preserved Ejection Fraction: a Genome-Wide Association Analysis of the Cardiovascular Health Study.

Little is known about genetics of heart failure with preserved ejection fraction (HFpEF) in part because of the many comorbidities in this population. To identify single-nucleotide polymorphisms (SNPs) associated with HFpEF, we analyzed phenotypic and genotypic data from the Cardiovascular Health Study, which profiled patients using a 50,000 SNP array. Results were explored using novel SNP- and gene-centric tools. We performed analyses to determine whether some SNPs were relevant only in certain phenotypes. Among 3804 patients, 7 clinical factors and 9 SNPs were significantly associated with HFpEF; the most notable of which was rs6996224, a SNP associated with transforming growth factor-beta receptor 3. Most SNPs were associated with HFpEF only in the absence of a clinical predictor. Significant SNPs represented genes involved in myocyte proliferation, transforming growth factor-beta/erbB signaling, and extracellular matrix formation. These findings suggest that genetic factors may be more important in some phenotypes than others.

Visual analysis of biological data-knowledge networks.

BACKGROUND: The interpretation of the results from genome-scale experiments is a challenging and important problem in contemporary biomedical research. Biological networks that integrate experimental results with existing knowledge from biomedical databases and published literature can provide a rich resource and powerful basis for hypothesizing about mechanistic explanations for observed gene-phenotype relationships. However, the size and density of such networks often impede their efficient exploration and understanding. RESULTS: We introduce a visual analytics approach that integrates interactive filtering of dense networks based on degree-of-interest functions with attribute-based layouts of the resulting subnetworks. The comparison of multiple subnetworks representing different analysis facets is facilitated through an interactive super-network that integrates brushing-and-linking techniques for highlighting components across networks. An implementation is freely available as a Cytoscape app. CONCLUSIONS: We demonstrate the utility of our approach through two case studies using a dataset that combines clinical data with high-throughput data for studying the effect of ß-blocker treatment on heart failure patients. Furthermore, we discuss our team-based iterative design and development process as well as the limitations and generalizability of our approach.

Peax: interactive visual analysis and exploration of complex clinical phenotype and gene expression association.

Increasing availability of high-dimensional clinical data, which improves the ability to define more specific phenotypes, as well as molecular data, which can elucidate disease mechanisms, is a driving force and at the same time a major challenge for translational and personalized medicine. Successful research in this field requires an approach that ties together specific disease and health expertise with understanding of molecular data through statistical methods. We present PEAX (Phenotype-Expression Association eXplorer), built upon open-source software, which integrates visual phenotype model definition with statistical testing of expression data presented concurrently in a web-browser. The integration of data and analysis tasks in a single tool allows clinical domain experts to obtain new insights directly through exploration of relationships between multivariate phenotype models and gene expression data, showing the effects of model definition and modification while also exploiting potential meaningful associations between phenotype and miRNA-mRNA regulatory relationships. We combine the web visualization capabilities of Shiny and D3 with the power and speed of R for backend statistical analysis, in order to abstract the scripting required for repetitive analysis of sub-phenotype association. We describe the motivation for PEAX, demonstrate its utility through a use case involving heart failure research, and discuss computational challenges and observations. We show that our visual web-based representations are well-suited for rapid exploration of phenotype and gene expression association, facilitating insight and discovery by domain experts.

Training the next generation of quantitative biologists in the era of big data.

The following sections are included: Workshop Focus, Workshop Contributions and References.

Ontologies in biological data visualization.

In computer science, an ontology is essentially a graph-based knowledge representation in which each node corresponds to a concept and each edge specifies a relation between two concepts. Ontological development in biology can serve as a focus to discuss the challenges and possible research directions for ontologies in visualization. The principle challenges are the dynamic and evolving nature of ontologies, the ever-present issue of scale, the diversity and richness of the relationships in ontologies, and the need to better understand the relationship between ontologies and the data analysis tasks scientists wish to support. Research directions include visualizing ontologies; visualizing semantically or ontologically annotated texts, documents, and corpora; automated generation of visualizations using ontologies; and visualizing ontological context to support search. Although this discussion uses issues of ontologies in biological data visualization as a springboard, these topics are of general relevance to visualization.

Scientists' sense making when hypothesizing about disease mechanisms from expression data and their needs for visualization support.

A common class of biomedical analysis is to explore expression data from high throughput experiments for the purpose of uncovering functional relationships that can lead to a hypothesis about mechanisms of a disease. We call this analysis expression driven, -omics hypothesizing. In it, scientists use interactive data visualizations and read deeply in the research literature. Little is known, however, about the actual flow of reasoning and behaviors (sense making) that scientists enact in this analysis, end-to-end. Understanding this flow is important because if bioinformatics tools are to be truly useful they must support it. Sense making models of visual analytics in other domains have been developed and used to inform the design of useful and usable tools. We believe they would be helpful in bioinformatics. To characterize the sense making involved in expression-driven, -omics hypothesizing, we conducted an in-depth observational study of one scientist as she engaged in this analysis over six months. From findings, we abstracted a preliminary sense making model. Here we describe its stages and suggest guidelines for developing visualization tools that we derived from this case. A single case cannot be generalized. But we offer our findings, sense making model and case-based tool guidelines as a first step toward increasing interest and further research in the bioinformatics field on scientists' analytical workflows and their implications for tool design.