Variational Bayes and the reduced dependence approximation for the autologistic model on an irregular grid with applications

Discrete Markov random field models provide a natural framework for representing images or spatial datasets. They model the spatial association present while providing a convenient Markovian dependency structure and strong edge-preservation properties. However, parameter estimation for discrete Markov random field models is difficult due to the complex form of the associated normalizing constant for the likelihood function. For large lattices, the reduced dependence approximation to the normalizing constant is based on the concept of performing computationally efficient and feasible forward recursions on smaller sublattices which are then suitably combined to estimate the constant for the whole lattice. We present an efficient computational extension of the forward recursion approach for the autologistic model to lattices that have an irregularly shaped boundary and which may contain regions with no data; these lattices are typical in applications. Consequently, we also extend the reduced dependence approximation to these scenarios enabling us to implement a practical and efficient non-simulation based approach for spatial data analysis within the variational Bayesian framework. The methodology is illustrated through application to simulated data and example images. The supplemental materials include our C++ source code for computing the approximate normalizing constant and simulation studies.

Phase III trial comparing paclitaxel poliglumex (CT-2103, PPX) in combination with carboplatin versus standard paclitaxel and carboplatin in the treatment of PS 2 patients with chemotherapy-naïve advanced non-small cell lung cancer

INTRODUCTION: Performance status (PS) 2 patients with non-small cell lung cancer (NSCLC) experience more toxicity, lower response rates, and shorter survival times than healthier patients treated with standard chemotherapy. Paclitaxel poliglumex (PPX), a macromolecule drug conjugate of paclitaxel and polyglutamic acid, reduces systemic exposure to peak concentrations of free paclitaxel and may lead to increased concentrations in tumors due to enhanced vascular permeability. METHODS: Chemotherapy-naive PS 2 patients with advanced NSCLC were randomized to receive carboplatin (area under the curve = 6) and either PPX (210 mg/m/10 min without routine steroid premedication) or paclitaxel (225 mg/m/3 h with standard premedication) every 3 weeks. The primary end point was overall survival. RESULTS: A total of 400 patients were enrolled. Alopecia, arthralgias/myalgias, and cardiac events were significantly less frequent with PPX/carboplatin, whereas grade ≥3 neutropenia and grade 3 neuropathy showed a trend of worsening. There was no significant difference in the incidence of hypersensitivity reactions despite the absence of routine premedication in the PPX arm. Overall survival was similar between treatment arms (hazard ratio, 0.97; log rank p = 0.769). Median and 1-year survival rates were 7.9 months and 31%, for PPX versus 8 months and 31% for paclitaxel. Disease control rates were 64% and 69% for PPX and paclitaxel, respectively. Time to progression was similar: 3.9 months for PPX/carboplatin versus 4.6 months for paclitaxel/carboplatin (p = 0.210). CONCLUSION: PPX/carboplatin failed to provide superior survival compared with paclitaxel/carboplatin in the first-line treatment of PS 2 patients with NSCLC, but the results with respect to progression-free survival and overall survival were comparable and the PPX regimen was more convenient. © 2008International Association for the Study of Lung Cancer.

PAC-Bayes-empirical-Bernstein inequality

We present PAC-Bayes-Empirical-Bernstein inequality. The inequality is based on combination of PAC-Bayesian bounding technique with Empirical Bernstein bound. It allows to take advantage of small empirical variance and is especially useful in regression. We show that when the empirical variance is significantly smaller than the empirical loss PAC-Bayes-Empirical-Bernstein inequality is significantly tighter than PAC-Bayes-kl inequality of Seeger (2002) and otherwise it is comparable. PAC-Bayes-Empirical-Bernstein inequality is an interesting example of application of PAC-Bayesian bounding technique to self-bounding functions. We provide empirical comparison of PAC-Bayes-Empirical-Bernstein inequality with PAC-Bayes-kl inequality on a synthetic example and several UCI datasets.

Pituitary volume and early treatment response in drug-naïve first-episode psychosis patients

BACKGROUND: An early response to antipsychotic treatment in patients with psychosis has been associated with a better course and outcome. However, factors that predict treatment response are not well understood. The onset of schizophrenia and related disorders has been associated with increased levels of stress and hyper-activation of the hypothalamic-pituitary-adrenal (HPA) axis. This study examined whether pituitary volume at the onset of psychosis may be a potential predictor of early treatment response in first-episode psychosis (FEP) patients. METHODS: We investigated the relationship between baseline pituitary volume and symptomatic treatment response over 12 weeks using mixed model analysis in a sample of 42 drug-naïve or early treated FEP patients who participated in a controlled dose-finding study of quetiapine fumarate. Logistic regression was used to examine predictors of treatment response. Pituitary volume was measured from magnetic resonance imaging scans that were obtained upon entry into the trial. RESULTS: Larger pituitary volume was associated with less improvement in overall psychotic symptoms (Brief Psychiatric Rating Scale (BPRS) P=0.031) and positive symptoms (BPRS positive symptom subscale P=0.010). Regardless of gender, patients with a pituitary volume at the 25th percentile (413 mm(3)) were approximately three times more likely to respond to treatment by week 12 than those at the 75th percentile (635 mm(3)) (odds ratio=3.07, CI: 0.90-10.48). CONCLUSION: The association of baseline pituitary volumes with early treatment response highlights the importance of the HPA axis in emerging psychosis. Potential implications for treatment strategies in early psychosis are discussed.

Transdimensional sequential Monte Carlo using variational Bayes — SMCVB

A new transdimensional Sequential Monte Carlo (SMC) algorithm called SM- CVB is proposed. In an SMC approach, a weighted sample of particles is generated from a sequence of probability distributions which ‘converge’ to the target distribution of interest, in this case a Bayesian posterior distri- bution. The approach is based on the use of variational Bayes to propose new particles at each iteration of the SMCVB algorithm in order to target the posterior more efficiently. The variational-Bayes-generated proposals are not limited to a fixed dimension. This means that the weighted particle sets that arise can have varying dimensions thereby allowing us the option to also estimate an appropriate dimension for the model. This novel algorithm is outlined within the context of finite mixture model estimation. This pro- vides a less computationally demanding alternative to using reversible jump Markov chain Monte Carlo kernels within an SMC approach. We illustrate these ideas in a simulated data analysis and in applications.

Combining Bayes Classification and Point Group Symmetry under Boolean Framework for Enhanced Protein Quaternary Structure Inference

Our ability to infer the protein quaternary structure automatically from atom and lattice information is inadequate, especially for weak complexes, and heteromeric quaternary structures. Several approaches exist, but they have limited performance. Here, we present a new scheme to infer protein quaternary structure from lattice and protein information, with all-around coverage for strong, weak and very weak affinity homomeric and heteromeric complexes. The scheme combines naive Bayes classifier and point group symmetry under Boolean framework to detect quaternary structures in crystal lattice. It consistently produces >= 90% coverage across diverse benchmarking data sets, including a notably superior 95% coverage for recognition heteromeric complexes, compared with 53% on the same data set by current state-of-the-art method. The detailed study of a limited number of prediction-failed cases offers interesting insights into the intriguing nature of protein contacts in lattice. The findings have implications for accurate inference of quaternary states of proteins, especially weak affinity complexes.