Patrick Breheny

COORDINATE DESCENT ALGORITHMS FOR NONCONVEX PENALIZED REGRESSION, WITH APPLICATIONS TO BIOLOGICAL FEATURE SELECTION

A number of variable selection methods have been proposed involving nonconvex penalty functions. These methods, which include the smoothly clipped absolute deviation (SCAD) penalty and the minimax concave penalty (MCP), have been demonstrated to have attractive theoretical properties, but model fitting is not a straightforward task, and the resulting solutions may be unstable. Here, we demonstrate the potential of coordinate descent algorithms for fitting these models, establishing theoretical convergence properties and demonstrating that they are significantly faster than competing approaches. In addition, we demonstrate the utility of convexity diagnostics to determine regions of the parameter space in which the objective function is locally convex, even though the penalty is not. Our simulation study and data examples indicate that nonconvex penalties like MCP and SCAD are worthwhile alternatives to the lasso in many applications. In particular, our numerical results suggest that MCP is the preferred approach among the three methods.

A Selective Review of Group Selection in High-Dimensional Models.

Grouping structures arise naturally in many statistical modeling problems. Several methods have been proposed for variable selection that respect grouping structure in variables. Examples include the group LASSO and several concave group selection methods. In this article, we give a selective review of group selection concerning methodological developments, theoretical properties and computational algorithms. We pay particular attention to group selection methods involving concave penalties. We address both group selection and bi-level selection methods. We describe several applications of these methods in nonparametric additive models, semiparametric regression, seemingly unrelated regressions, genomic data analysis and genome wide association studies. We also highlight some issues that require further study.

Surface sensing in Vibrio parahaemolyticus triggers a programme of gene expression that promotes colonization and virulence

Vibrio parahaemolyticus senses surfaces via impeded rotation of its polar flagellum. We have exploited this surface‐sensing mechanism to trick the organism into thinking it is on a surface when it is growing in liquid. This facilitated studies of global gene expression in a way that avoided many of the complications of surface‐to‐liquid comparisons, and illuminated ∼ 70 genes that respond to surface sensing per se. Almost all are surface‐induced (not repressed) and encode swarming motility proteins, virulence factors or sensory enzymes involved with chemoreception and c‐di‐GMP signalling. Follow‐up studies were performed to place the surface‐responsive genes in a regulatory hierarchy. Mapping the hierarchy revealed two surprises about LafK, a transcriptional activator that until now has been considered to be the master regulator for the lateral flagellar system. First, LafK controls a more diverse set of genes than previously appreciated. Second, some laf genes are not under LafK control, which means LafK is not the master regulator after all. Additional experiments motivated by the transcriptome analyses revealed that growth on a surface lowers c‐di‐GMP levels and enhances cytotoxicity. Thus, we demonstrate that V. parahaemolyticus can invoke a programme of gene control upon encountering a surface and the specific identities of the surface‐responsive genes are pertinent to colonization and pathogenesis.

Group descent algorithms for nonconvex penalized linear and logistic regression models with grouped predictors

Penalized regression is an attractive framework for variable selection problems. Often, variables possess a grouping structure, and the relevant selection problem is that of selecting groups, not individual variables. The group lasso has been proposed as a way of extending the ideas of the lasso to the problem of group selection. Nonconvex penalties such as SCAD and MCP have been proposed and shown to have several advantages over the lasso; these penalties may also be extended to the group selection problem, giving rise to group SCAD and group MCP methods. Here, we describe algorithms for fitting these models stably and efficiently. In addition, we present simulation results and real data examples comparing and contrasting the statistical properties of these methods.

Genomics of mature and immature olfactory sensory neurons.

The continuous replacement of neurons in the olfactory epithelium provides an advantageous model for investigating neuronal differentiation and maturation. By calculating the relative enrichment of every mRNA detected in samples of mature mouse olfactory sensory neurons (OSNs), immature OSNs, and the residual population of neighboring cell types, and then comparing these ratios against the known expression patterns of >300 genes, enrichment criteria that accurately predicted the OSN expression patterns of nearly all genes were determined. We identified 847 immature OSN‐specific and 691 mature OSN‐specific genes. The control of gene expression by chromatin modification and transcription factors, and neurite growth, protein transport, RNA processing, cholesterol biosynthesis, and apoptosis via death domain receptors, were overrepresented biological processes in immature OSNs. Ion transport (ion channels), presynaptic functions, and cilia‐specific processes were overrepresented in mature OSNs. Processes overrepresented among the genes expressed by all OSNs were protein and ion transport, ER overload response, protein catabolism, and the electron transport chain. To more accurately represent gradations in mRNA abundance and identify all genes expressed in each cell type, classification methods were used to produce probabilities of expression in each cell type for every gene. These probabilities, which identified 9,300 genes expressed in OSNs, were 96% accurate at identifying genes expressed in OSNs and 86% accurate at discriminating genes specific to mature and immature OSNs. This OSN gene database not only predicts the genes responsible for the major biological processes active in OSNs, but also identifies thousands of never before studied genes that support OSN phenotypes. J. Comp. Neurol. 520:2608–2629, 2012.

Primary radiation therapy for medically inoperable patients with clinical stage I and II endometrial carcinoma

OBJECTIVE To determine the outcomes associated with primary radiation therapy for medically inoperable, clinical stage I and II, endometrial adenocarcinoma (EAC). METHODS A multi-institution, retrospective chart review from January 1997 to January 2009 was performed. Overall survival (OS), disease-specific survival (DSS), progression-free survival (PFS) and time to progression (TTP) were assessed using the Kaplan-Meier method. Disease-specific survival was analyzed using a competing risks approach. RESULTS Seventy-four patients were evaluable. The median age and BMI were 65 years (range 36-92 years) and 46 kg/m(2) (range 23-111 kg/m(2)), respectively. 85.1% had severe systemic disease, most frequently cardiopulmonary risk and morbid obesity. With a mean follow-up of 31 months, 13 patients (17.6%) experienced a recurrence. The median PFS and OS were 43.5 months and 47.2 months, respectively. Overall, 35 women died, including 4 women who died of unknown cause. Of the remaining 31 women, 7 patients (9.5%) died of disease, while 24 died of other causes (32.4%). The hazard ratio comparing the risk of death due to other causes to the risk of death due to disease was 3.4 (95% CI 1.4-9.4, p=0.003). Among patients who are alive three years after diagnosis, 14% recurred and the conditional recurrence estimate did not exceed 16%. CONCLUSIONS Primary radiation therapy for clinical stage I and II EAC is a feasible option for medically inoperable patients and provides disease control, with fewer than 16% of surviving patients experiencing recurrence.

Penalized Multimarker vs. Single-Marker Regression Methods for Genome-Wide Association Studies of Quantitative Traits

The data from genome-wide association studies (GWAS) in humans are still predominantly analyzed using single-marker association methods. As an alternative to single-marker analysis (SMA), all or subsets of markers can be tested simultaneously. This approach requires a form of penalized regression (PR) as the number of SNPs is much larger than the sample size. Here we review PR methods in the context of GWAS, extend them to perform penalty parameter and SNP selection by false discovery rate (FDR) control, and assess their performance in comparison with SMA. PR methods were compared with SMA, using realistically simulated GWAS data with a continuous phenotype and real data. Based on these comparisons our analytic FDR criterion may currently be the best approach to SNP selection using PR for GWAS. We found that PR with FDR control provides substantially more power than SMA with genome-wide type-I error control but somewhat less power than SMA with Benjamini–Hochberg FDR control (SMA-BH). PR with FDR-based penalty parameter selection controlled the FDR somewhat conservatively while SMA-BH may not achieve FDR control in all situations. Differences among PR methods seem quite small when the focus is on SNP selection with FDR control. Incorporating linkage disequilibrium into the penalization by adapting penalties developed for covariates measured on graphs can improve power but also generate more false positives or wider regions for follow-up. We recommend the elastic net with a mixing weight for the Lasso penalty near 0.5 as the best method.

In Vivo Identification of Eugenol-Responsive and Muscone-Responsive Mouse Odorant Receptors

Our understanding of mammalian olfactory coding has been impeded by the paucity of information about the odorant receptors (ORs) that respond to a given odorant ligand in awake, freely behaving animals. Identifying the ORs that respond in vivo to a given odorant ligand from among the ∼1100 ORs in mice is intrinsically challenging but critical for our understanding of olfactory coding at the periphery. Here, we report an in vivo assay that is based on a novel gene-targeted mouse strain, S100a5–tauGFP, in which a fluorescent reporter selectively marks olfactory sensory neurons that have been activated recently in vivo. Because each olfactory sensory neuron expresses a single OR gene, multiple ORs responding to a given odorant ligand can be identified simultaneously by capturing the population of activated olfactory sensory neurons and using expression profiling methods to screen the repertoire of mouse OR genes. We used this in vivo assay to re-identify known eugenol- and muscone-responsive mouse ORs. We identified additional ORs responsive to eugenol or muscone. Heterologous expression assays confirmed nine eugenol-responsive ORs (Olfr73, Olfr178, Olfr432, Olfr610, Olfr958, Olfr960, Olfr961, Olfr913, and Olfr1234) and four muscone-responsive ORs (Olfr74, Olfr235, Olfr816, and Olfr1440). We found that the human ortholog of Olfr235 and Olfr1440 responds to macrocyclic ketone and lactone musk odorants but not to polycyclic musk odorants or a macrocyclic diester musk odorant. This novel assay, called the Kentucky in vivo odorant ligand–receptor assay, should facilitate the in vivo identification of mouse ORs for a given odorant ligand of interest.

Morphine Versus Clonidine for Neonatal Abstinence Syndrome

OBJECTIVE: The study goal was to determine whether clonidine treatment of neonatal abstinence syndrome (NAS) would result in a better neurobehavioral performance compared with morphine. METHODS: This pilot study prospectively enrolled infants ≥35 weeks’ gestational age admitted for treatment of NAS. After informed consent was obtained, infants were randomized to receive morphine (0.4 mg/kg per day) or clonidine (5 μg/kg per day) divided into 8 doses. A 25% dose escalation every 24 hours was possible per protocol (maximum of 1 mg/kg per day for morphine and 12 μg/kg per day for clonidine). After control of symptoms, the dose was tapered by 10% every other day. Clinical staff monitored infants by using Finnegan scoring. Masked research staff administered the NICU Network Neurobehavioral Scale (NNNS) at 1 week and at 2 to 4 weeks after initiation of treatment and the Bayley Scales III, and Preschool Language Scale IV, at 1-year adjusted age. Analyses included descriptive statistics, repeated measures analysis of variance, and Wilcoxon tests. RESULTS: Infants treated with morphine (n = 15) versus clonidine (n = 16) did not differ in birth weight or age at treatment. Treatment duration was significantly longer for morphine (median 39 days) than for clonidine (median 28 days; P = .02). NNNS summary scores improved significantly with clonidine but not with morphine. On subsequent assessment, those receiving clonidine had lower height of arousal and excitability (P < .05). One-year motor, cognitive, and language scores did not differ between groups. CONCLUSIONS: Clonidine may be a favorable alternative to morphine as a single-drug therapy for NAS. A multicenter randomized trial is warranted.

Impact of Obesity on the Risk of Venous Thromboembolism in an Inpatient Pediatric Population

Introduction: The incidence of venous thromboembolism in children has increased significantly over the past 20 years. Over the same period of time, there was an increase in the prevalence of obesity in the pediatric population. Obesity is a known risk factor for VTE in adults, but little information is available in children. Methods: This study evaluates the relation between obesity and VTE using a retrospective, case–control design, comparing the body mass index (BMI) of patients admitted with a diagnosis of VTE versus patients admitted with other diagnoses, at a single institution, between 2007 and 2011. Results: We studied 48 inpatients diagnosed with deep venous thrombosis or pulmonary embolism and a control group of 274 age and gender matched patients admitted with other diagnoses. We found obese patients (BMI > 95th percentile) to have significantly higher risk of VTE (odds ratio 2.1, with 95% CI 1.1–4.2) than patients of normal weight (BMI < 85th percentile). Overweight patients (BMI 85th–95th percentile) did not demonstrate a significant change in risk. Most of the VTE patients were adolescents and the majority of them had other identifiable risk factors for thrombosis. Conclusion: This study establishes a correlation between obesity and VTE in a group of hospitalized children, showing a risk for VTE in obese children similar to the one described in much larger adult cohorts.