Optimization of Genomic Classifiers for Clinical Deployment: Evaluation of Bayesian Optimization to Select Predictive Models of Acute Infection and In-Hospital Mortality.

Acute infection, if not rapidly and accurately detected, can lead to sepsis, organ failure and even death. Current detection of acute infection as well as assessment of a patient's severity of illness are imperfect. Characterization of a patient's immune response by quantifying expression levels of specific genes from blood represents a potentially more timely and precise means of accomplishing both tasks. Machine learning methods provide a platform to leverage this host response for development of deployment-ready classification models. Prioritization of promising classifiers is dependent, in part, on hyperparameter optimization for which a number of approaches including grid search, random sampling and Bayesian optimization have been shown to be effective. We compare HO approaches for the development of diagnostic classifiers of acute infection and in-hospital mortality from gene expression of 29 diagnostic markers. We take a deployment-centered approach to our comprehensive analysis, accounting for heterogeneity in our multi-study patient cohort with our choices of dataset partitioning and hyperparameter optimization objective as well as assessing selected classifiers in external (as well as internal) validation. We find that classifiers selected by Bayesian optimization for in-hospital mortality can outperform those selected by grid search or random sampling. However, in contrast to previous research: 1) Bayesian optimization is not more efficient in selecting classifiers in all instances compared to grid search or random sampling-based methods and 2) we note marginal gains in classifier performance in only specific circumstances when using a common variant of Bayesian optimization (i.e. automatic relevance determination). Our analysis highlights the need for further practical, deployment-centered benchmarking of HO approaches in the healthcare context.

A generalizable 29-mRNA neural-network classifier for acute bacterial and viral infections.

Improved identification of bacterial and viral infections would reduce morbidity from sepsis, reduce antibiotic overuse, and lower healthcare costs. Here, we develop a generalizable host-gene-expression-based classifier for acute bacterial and viral infections. We use training data (N = 1069) from 18 retrospective transcriptomic studies. Using only 29 preselected host mRNAs, we train a neural-network classifier with a bacterial-vs-other area under the receiver-operating characteristic curve (AUROC) 0.92 (95% CI 0.90-0.93) and a viral-vs-other AUROC 0.92 (95% CI 0.90-0.93). We then apply this classifier, inflammatix-bacterial-viral-noninfected-version 1 (IMX-BVN-1), without retraining, to an independent cohort (N = 163). In this cohort, IMX-BVN-1 AUROCs are: bacterial-vs.-other 0.86 (95% CI 0.77-0.93), and viral-vs.-other 0.85 (95% CI 0.76-0.93). In patients enrolled within 36 h of hospital admission (N = 70), IMX-BVN-1 AUROCs are: bacterial-vs.-other 0.92 (95% CI 0.83-0.99), and viral-vs.-other 0.91 (95% CI 0.82-0.98). With further study, IMX-BVN-1 could provide a tool for assessing patients with suspected infection and sepsis at hospital admission.

Flexible, cluster-based analysis of the electronic medical record of sepsis with composite mixture models.

The widespread adoption of electronic medical records (EMRs) in healthcare has provided vast new amounts of data for statistical machine learning researchers in their efforts to model and predict patient health status, potentially enabling novel advances in treatment. In the case of sepsis, a debilitating, dysregulated host response to infection, extracting subtle, uncataloged clinical phenotypes from the EMR with statistical machine learning methods has the potential to impact patient diagnosis and treatment early in the course of their hospitalization. However, there are significant barriers that must be overcome to extract these insights from EMR data. First, EMR datasets consist of both static and dynamic observations of discrete and continuous-valued variables, many of which may be missing, precluding the application of standard multivariate analysis techniques. Second, clinical populations observed via EMRs and relevant to the study and management of conditions like sepsis are often heterogeneous; properly accounting for this heterogeneity is critical. Here, we describe an unsupervised, probabilistic framework called a composite mixture model that can simultaneously accommodate the wide variety of observations frequently observed in EMR datasets, characterize heterogeneous clinical populations, and handle missing observations. We demonstrate the efficacy of our approach on a large-scale sepsis cohort, developing novel techniques built on our model-based clusters to track patient mortality risk over time and identify physiological trends and distinct subgroups of the dataset associated with elevated risk of mortality during hospitalization.

Size-Dependent Expression of the Mitotic Activator Cdc25 Suggests a Mechanism of Size Control in Fission Yeast.

Proper cell size is essential for cellular function. Nonetheless, despite more than 100 years of work on the subject, the mechanisms that maintain cell-size homeostasis are largely mysterious [1]. Cells in growing populations maintain cell size within a narrow range by coordinating growth and division. Bacterial and eukaryotic cells both demonstrate homeostatic size control, which maintains population-level variation in cell size within a certain range and returns the population average to that range if it is perturbed [1, 2]. Recent work has proposed two different strategies for size control: budding yeast has been proposed to use an inhibitor-dilution strategy to regulate size at the G1/S transition [3], whereas bacteria appear to use an adder strategy, in which a fixed amount of growth each generation causes cell size to converge on a stable average [4-6]. Here we present evidence that cell size in the fission yeast Schizosaccharomyces pombe is regulated by a third strategy: the size-dependent expression of the mitotic activator Cdc25. cdc25 transcript levels are regulated such that smaller cells express less Cdc25 and larger cells express more Cdc25, creating an increasing concentration of Cdc25 as cells grow and providing a mechanism for cells to trigger cell division when they reach a threshold concentration of Cdc25. Because regulation of mitotic entry by Cdc25 is well conserved, this mechanism may provide a widespread solution to the problem of size control in eukaryotes.

A comprehensive multi-omics approach uncovers adaptations for growth and survival of Pseudomonas aeruginosa on n-alkanes.

BACKGROUND: Examination of complex biological systems has long been achieved through methodical investigation of the system's individual components. While informative, this strategy often leads to inappropriate conclusions about the system as a whole. With the advent of high-throughput "omic" technologies, however, researchers can now simultaneously analyze an entire system at the level of molecule (DNA, RNA, protein, metabolite) and process (transcription, translation, enzyme catalysis). This strategy reduces the likelihood of improper conclusions, provides a framework for elucidation of genotype-phenotype relationships, and brings finer resolution to comparative genomic experiments. Here, we apply a multi-omic approach to analyze the gene expression profiles of two closely related Pseudomonas aeruginosa strains grown in n-alkanes or glycerol. RESULTS: The environmental P. aeruginosa isolate ATCC 33988 consumed medium-length (C10-C16) n-alkanes more rapidly than the laboratory strain PAO1, despite high genome sequence identity (average nucleotide identity >99%). Our data shows that ATCC 33988 induces a characteristic set of genes at the transcriptional, translational and post-translational levels during growth on alkanes, many of which differ from those expressed by PAO1. Of particular interest was the lack of expression from the rhl operon of the quorum sensing (QS) system, resulting in no measurable rhamnolipid production by ATCC 33988. Further examination showed that ATCC 33988 lacked the entire lasI/lasR arm of the QS response. Instead of promoting expression of QS genes, ATCC 33988 up-regulates a small subset of its genome, including operons responsible for specific alkaline proteases and sphingosine metabolism. CONCLUSION: This work represents the first time results from RNA-seq, microarray, ribosome footprinting, proteomics, and small molecule LC-MS experiments have been integrated to compare gene expression in bacteria. Together, these data provide insights as to why strain ATCC 33988 is better adapted for growth and survival on n-alkanes.

Characterization of dependencies between growth and division in budding yeast.

Cell growth and division are processes vital to the proliferation and development of life. Coordination between these two processes has been recognized for decades in a variety of organisms. In the budding yeast Saccharomyces cerevisiae, this coordination or 'size control' appears as an inverse correlation between cell size and the rate of cell-cycle progression, routinely observed in G1 prior to cell division commitment. Beyond this point, cells are presumed to complete S/G2/M at similar rates and in a size-independent manner. As such, studies of dependence between growth and division have focused on G1 Moreover, in unicellular organisms, coordination between growth and division has commonly been analysed within the cycle of a single cell without accounting for correlations in growth and division characteristics between cycles of related cells. In a comprehensive analysis of three published time-lapse microscopy datasets, we analyse both intra- and inter-cycle dependencies between growth and division, revisiting assumptions about the coordination between these two processes. Interestingly, we find evidence (i) that S/G2/M durations are systematically longer in daughters than in mothers, (ii) of dependencies between S/G2/M and size at budding that echo the classical G1 dependencies, and (iii) in contrast with recent bacterial studies, of negative dependencies between size at birth and size accumulated during the cell cycle. In addition, we develop a novel hierarchical model to uncover inter-cycle dependencies, and we find evidence for such dependencies in cells growing in sugar-poor environments. Our analysis highlights the need for experimentalists and modellers to account for new sources of cell-to-cell variation in growth and division, and our model provides a formal statistical framework for the continued study of dependencies between biological processes.

Cyclin-dependent kinases are regulators and effectors of oscillations driven by a transcription factor network.

During embryonic cell cycles, B-cyclin-CDKs function as the core component of an autonomous oscillator. Current models for the cell-cycle oscillator in nonembryonic cells are slightly more complex, incorporating multiple G1, S phase, and mitotic cyclin-CDK complexes. However, periodic events persist in yeast cells lacking all S phase and mitotic B-cyclin genes, challenging the assertion that cyclin-CDK complexes are essential for oscillations. These and other results led to the proposal that a network of sequentially activated transcription factors functions as an underlying cell-cycle oscillator. Here we examine the individual contributions of a transcription factor network and cyclin-CDKs to the maintenance of cell-cycle oscillations. Our findings suggest that while cyclin-CDKs are not required for oscillations, they do contribute to oscillation robustness. A model emerges in which cyclin expression (thereby, CDK activity) is entrained to an autonomous transcriptional oscillator. CDKs then modulate oscillator function and serve as effectors of the oscillator.

A generalized model for multi-marker analysis of cell cycle progression in synchrony experiments.

MOTIVATION: To advance understanding of eukaryotic cell division, it is important to observe the process precisely. To this end, researchers monitor changes in dividing cells as they traverse the cell cycle, with the presence or absence of morphological or genetic markers indicating a cell's position in a particular interval of the cell cycle. A wide variety of marker data is available, including information-rich cellular imaging data. However, few formal statistical methods have been developed to use these valuable data sources in estimating how a population of cells progresses through the cell cycle. Furthermore, existing methods are designed to handle only a single binary marker of cell cycle progression at a time. Consequently, they cannot facilitate comparison of experiments involving different sets of markers. RESULTS: Here, we develop a new sampling model to accommodate an arbitrary number of different binary markers that characterize the progression of a population of dividing cells along a branching process. We engineer a strain of Saccharomyces cerevisiae with fluorescently labeled markers of cell cycle progression, and apply our new model to two image datasets we collected from the strain, as well as an independent dataset of different markers. We use our model to estimate the duration of post-cytokinetic attachment between a S.cerevisiae mother and daughter cell. The Java implementation is fast and extensible, and includes a graphical user interface. Our model provides a powerful and flexible cell cycle analysis tool, suitable to any type or combination of binary markers. AVAILABILITY: The software is available from: http://www.cs.duke.edu/~amink/software/cloccs/. CONTACT: michael.mayhew@duke.edu; amink@cs.duke.edu.