Scientists Invent New Hypotheses, Do Brains?

How are new Bayesian hypotheses generated within the framework of predictive processing? This explanatory framework purports to provide a unified, systematic explanation of cognition by appealing to Bayes rule and hierarchical Bayesian machinery alone. Given that the generation of new hypotheses is fundamental to Bayesian inference, the predictive processing framework faces an important challenge in this regard. By examining several cognitive-level and neurobiological architecture-inspired models of hypothesis generation, we argue that there is an essential difference between the two types of models. Cognitive-level models do not specify how they can be implemented in brains and include structures and assumptions that are external to the predictive processing framework. By contrast, neurobiological architecture-inspired models, which aim to better resemble brain processes, fail to explain important capacities of cognition, such as categorization and few-shot learning. The "scaling-up" challenge for proponents of predictive processing is to explain the relationship between these two types of models using only the theoretical and conceptual machinery of Bayesian inference.

Interpretation and Analysis of Individual Diagnostic Tests and Performance

Diagnostic tests are performed daily by bovine practitioners at the individual and population level. At the individual level, they help not only for making a diagnosis, but can also serve to rule in or rule out a specific condition, monitor treatment response, establish a prognosis, or to determine infection status. Performing an individual diagnostic test is technical; however, its interpretation and contextualization requires medical and epidemiologic skills that veterinary practitioners are able to master. This article shows the added value of the context of test prescription and correct interpretation highlighting the central role of the veterinary practitioner.

Novel Meta-Learning Techniques for the Multiclass Image Classification Problem

Multiclass image classification is a complex task that has been thoroughly investigated in the past. Decomposition-based strategies are commonly employed to address it. Typically, these methods divide the original problem into smaller, potentially simpler problems, allowing the application of numerous well-established learning algorithms that may not apply directly to the original task. This work focuses on the efficiency of decomposition-based methods and proposes several improvements to the meta-learning level. In this paper, four methods for optimizing the ensemble phase of multiclass classification are introduced. The first demonstrates that employing a mixture of experts scheme can drastically reduce the number of operations in the training phase by eliminating redundant learning processes in decomposition-based techniques for multiclass problems. The second technique for combining learner-based outcomes relies on Bayes' theorem. Combining the Bayes rule with arbitrary decompositions reduces training complexity relative to the number of classifiers even further. Two additional methods are also proposed for increasing the final classification accuracy by decomposing the initial task into smaller ones and ensembling the output of the base learners along with that of a multiclass classifier. Finally, the proposed novel meta-learning techniques are evaluated on four distinct datasets of varying classification difficulty. In every case, the proposed methods present a substantial accuracy improvement over existing traditional image classification techniques.

Knowledge and the reliability of constructive memory

Memory is constructive, but that does not mean it is unreliable. When people remember the events of their lives they depend on knowledge, some of which is in the form of scripts or schemata. Schematic information encodes typical patterns in events, and for this reason schemata often contribute veridical features to memory reconstruction. This process can be thought of in Bayesian terms, as incorporating prior probabilities based on recurring patterns in experience. It also can be thought of in terms of statistical regression, such that information from knowledge is combined with information from episodic traces to reconstruct a best estimate of what happened.

COMPASS: Computations for Orientation and Motion Perception in Altered Sensorimotor States

Reliable perception of self-motion and orientation requires the central nervous system (CNS) to adapt to changing environments, stimuli, and sensory organ function. The proposed computations required of neural systems for this adaptation process remain conceptual, limiting our understanding and ability to quantitatively predict adaptation and mitigate any resulting impairment prior to completing adaptation. Here, we have implemented a computational model of the internal calculations involved in the orientation perception system’s adaptation to changes in the magnitude of gravity. In summary, we propose that the CNS considers parallel, alternative hypotheses of the parameter of interest (in this case, the CNS’s internal estimate of the magnitude of gravity) and uses the associated sensory conflict signals (i.e., difference between sensory measurements and the expectation of them) to sequentially update the posterior probability of each hypothesis using Bayes rule. Over time, an updated central estimate of the internal magnitude of gravity emerges from the posterior probability distribution, which is then used to process sensory information and produce perceptions of self-motion and orientation. We have implemented these hypotheses in a computational model and performed various simulations to demonstrate quantitative model predictions of adaptation of the orientation perception system to changes in the magnitude of gravity, similar to those experienced by astronauts during space exploration missions. These model predictions serve as quantitative hypotheses to inspire future experimental assessments.

An incremental learning approach to automatically recognize pulmonary diseases from the multi-vendor chest radiographs

The human respiratory network is a vital system that provides oxygen supply and nourishment to the whole body. Pulmonary diseases can cause severe respiratory problems, leading to sudden death if not treated timely. Many researchers have utilized deep learning systems (in both transfer learning and fine-tuning modes) to diagnose pulmonary disorders using chest X-rays (CXRs). However, such systems require exhaustive training efforts on large-scale (and well-annotated) data to effectively diagnose chest abnormalities (at the inference stage). Furthermore, procuring such large-scale data (in a clinical setting) is often infeasible and impractical, especially for rare diseases. With the recent advances in incremental learning, researchers have periodically tuned deep neural networks to learn different classification tasks with few training examples. Although, such systems can resist catastrophic forgetting, they treat the knowledge representations (which the network learns periodically) independently of each other, and this limits their classification performance. Also, to the best of our knowledge, there is no incremental learning-driven image diagnostic framework (to date) that is specifically designed to screen pulmonary disorders from the CXRs. To address this, we present a novel framework that can learn to screen different chest abnormalities incrementally (via few-shot training). In addition to this, the proposed framework is penalized through an incremental learning loss function that infers Bayesian theory to recognize structural and semantic inter-dependencies between incrementally learned knowledge representations to diagnose the pulmonary diseases effectively (at the inference stage), regardless of the scanner specifications. We tested the proposed framework on five public CXR datasets containing different chest abnormalities, where it achieved an accuracy of 0.8405 and the F1 score of 0.8303, outperforming various state-of-the-art incremental learning schemes. It also achieved a highly competitive performance compared to the conventional fine-tuning (transfer learning) approaches while significantly reducing the training and computational requirements.

Decomposing the Apoptosis Pathway Into Biologically Interpretable Principal Components

Principal component analysis (PCA) is one of the most common techniques in the analysis of biological data sets, but applying PCA raises 2 challenges. First, one must determine the number of significant principal components (PCs). Second, because each PC is a linear combination of genes, it rarely has a biological interpretation. Existing methods to determine the number of PCs are either subjective or computationally extensive. We review several methods and describe a new R package, PCDimension, that implements additional methods, the most important being an algorithm that extends and automates a graphical Bayesian method. Using simulations, we compared the methods. Our newly automated procedure is competitive with the best methods when considering both accuracy and speed and is the most accurate when the number of objects is small compared with the number of attributes. We applied the method to a proteomics data set from patients with acute myeloid leukemia. Proteins in the apoptosis pathway could be explained using 6 PCs. By clustering the proteins in PC space, we were able to replace the PCs by 6 "biological components," 3 of which could be immediately interpreted from the current literature. We expect this approach combining PCA with clustering to be widely applicable.

Determining the likely place of HIV acquisition for migrants in Europe combining subject-specific information and biomarkers data

In most HIV-positive individuals, infection time is only known to lie between the time an individual started being at risk for HIV and diagnosis time. However, a more accurate estimate of infection time is very important in certain cases. For example, one of the objectives of the Advancing Migrant Access to Health Services in Europe (aMASE) study was to determine if HIV-positive migrants, diagnosed in Europe, were infected pre- or post-migration. We propose a method to derive subject-specific estimates of unknown infection times using information from HIV biomarkers' measurements, demographic, clinical, and behavioral data. We assume that CD4 cell count (CD4) and HIV-RNA viral load trends after HIV infection follow a bivariate linear mixed model. Using post-diagnosis CD4 and viral load measurements and applying the Bayes' rule, we derived the posterior distribution of the HIV infection time, whereas the prior distribution was informed by AIDS status at diagnosis and behavioral data. Parameters of the CD4-viral load and time-to-AIDS models were estimated using data from a large study of individuals with known HIV infection times (CASCADE). Simulations showed substantial predictive ability (e.g. 84% of the infections were correctly classified as pre- or post-migration). Application to the aMASE study (n = 2009) showed that 47% of African migrants and 67% to 72% of migrants from other regions were most likely infected post-migration. Applying a Bayesian method based on bivariate modeling of CD4 and viral load, and subject-specific information, we found that the majority of HIV-positive migrants in aMASE were most likely infected after their migration to Europe.

Incorporating breeding abundance into spatial assignments on continuous surfaces

Determining the geographic connections between breeding and nonbreeding populations, termed migratory connectivity, is critical to advancing our understanding of the ecology and conservation of migratory species. Assignment models based on stable isotopes historically have been an important tool for studying migratory connectivity of small-bodied species, but the low resolution of these assignments has generated interest into combining isotopes with other sources in information. Abundance is one of the most appealing data sources to include in isotope-based assignments, but there are currently no statistical methods or guidelines for optimizing the contribution of stable isotopes and abundance for inferring migratory connectivity. Using known-origin stable-hydrogen isotope samples of six Neotropical migratory bird species, we rigorously assessed the performance of assignment models that differentially weight the contribution of the isotope and abundance data. For two species with adequate sample sizes, we used Pareto optimality to determine the set of models that simultaneously minimized both assignment error rate and assignment area. We then assessed the ability of the top models from these two species to improve assignments of the remaining four species compared to assignments based on isotopes alone. We show that the increased precision of models that include abundance is often offset by a large increase in assignment error. However, models that optimally weigh the abundance data relative to the isotope data can result in higher precision and, in some cases, lower error than models based on isotopes alone. The top models, however, depended on the distribution of relative breeding abundance, with patchier distributions requiring stronger downweighting of abundance, and we present general guidelines for future studies. These results confirm that breeding abundance can be an important source of information for studies investigating broad-scale movements of migratory birds and potentially other taxa.

Probability of an Autism Diagnosis by Gestational Age. NEWBORN AND INFANT NURSING REVIEWS : NAINR 2016;16:322-326. [PMID: 28989329 PMCID: PMC5627777 DOI: 10.1053/j.nainr.2016.09.019]

Early preterm infants (EPT) (<33 6/7 weeks) are at increased risk for autism spectrum disorders (ASDs) but prevalence estimates vary widely across studies. Furthermore, there are very few studies addressing the association between late preterm (LPT) births (34-36 6/7 weeks) and ASDs. To address the question of whether LPT infants carry the same risk for ASDs as full-term infants, this study aimed to estimate the relative probability of an ASD diagnosis using Bayes rule. A retrospective cohort analysis of 406 children was undertaken to look at gestational age, ASDs, and birth history. The application of Bayes rule was used, given that there is not sufficient information about the joint probabilities related to prematurity and autism. Using the estimated gestational age proportions within ASD diagnosis, plus national estimates of ASDs, probabilities for ASDs within a given gestational age were calculated. Among these 406 children with ASDs, 6.7% were EPT and 10.6% were LPT. In comparison to full term, EPT children are at 1.9 multiplicative increase in risk (95% CI [1.3, 2.5]). While the probability of ASDs for LPT children was higher than that for term, the estimated relative risk of the LPT infants was not statistically significant (95% CI [0.9, 1.5]). EPT infants were significantly more likely to be diagnosed with ASDs compared to their term peers. While the relative probability of ASD diagnosis among children born LPT was not statistically significant in this limited sample, the results indicate a possible elevated risk. A larger cohort is needed to adequately estimate this risk.

Bayesian analysis of high-resolution ultrasonography and guided fine needle aspiration cytology in diagnosis of palpable thyroid nodules

INTRODUCTION

To evaluate diagnostic accuracy of high-resolution ultrasonography in differentiation of benign and malignant thyroid nodules in comparison to results of guided fine needle aspiration cytology based on the Bayes rule.

OBJECTIVE

To assess the validity of ultrasonography results of thyroid nodules in comparison to guided fine needle aspiration cytology findings.

METHODS

This study was done on randomly chosen 80 patients presented with palpable thyroid nodules, undergone real-time sonographic evaluation of thyroid nodules to characterize features, internal consistency, margins, echotexture, calcification, peripheral lucent halo and vascularity. Ultrasonography guided fine needle aspiration cytology studies of thyroid nodules were done.

RESULTS

Palpable thyroid nodules were highly prevalent in fourth and fifth decades of life with female-male ratio, 4:1. Solid internal consistency was demonstrated by 75% malignant nodules. Hypoechogenicity and intra-nodular micro-calcifications were observed in 92% malignant nodules; 83% malignant nodules had intra-nodular vascularity and absence of peripheral halo. The pre-test prevalence of malignant nodules in the targeted population was 17.5%. As type I error, 2.5% false-positive cases and as type II error, 5.0% false-negative cases were detected. Values of sensitivity and specificity of the ultrasonography test were 71.43 and 96.97%, respectively.

CONCLUSION

Malignant thyroid nodules demonstrated ultrasonography characteristics of hypoechoic texture, intra-nodular micro-calcifications, solid consistency, internal vascularity and absence of peripheral halo. The ultrasonography test has 92.5% diagnostic accuracy to differentiate malignant from benign lesions in comparison to the gold standard fine needle aspiration cytology test.

Combining statistical inference and decisions in ecology

Statistical decision theory (SDT) is a sub-field of decision theory that formally incorporates statistical investigation into a decision-theoretic framework to account for uncertainties in a decision problem. SDT provides a unifying analysis of three types of information: statistical results from a data set, knowledge of the consequences of potential choices (i.e., loss), and prior beliefs about a system. SDT links the theoretical development of a large body of statistical methods, including point estimation, hypothesis testing, and confidence interval estimation. The theory and application of SDT have mainly been developed and published in the fields of mathematics, statistics, operations research, and other decision sciences, but have had limited exposure in ecology. Thus, we provide an introduction to SDT for ecologists and describe its utility for linking the conventionally separate tasks of statistical investigation and decision making in a single framework. We describe the basic framework of both Bayesian and frequentist SDT, its traditional use in statistics, and discuss its application to decision problems that occur in ecology. We demonstrate SDT with two types of decisions: Bayesian point estimation and an applied management problem of selecting a prescribed fire rotation for managing a grassland bird species. Central to SDT, and decision theory in general, are loss functions. Thus, we also provide basic guidance and references for constructing loss functions for an SDT problem.

A Bayesian methodology for detecting targeted genes under two related experiments

Many gene expression data are based on two experiments where the gene expressions of the targeted genes under both experiments are correlated. We consider problems in which objectives are to find genes that are simultaneously upregulated/downregulated under both experiments. A Bayesian methodology is proposed based on directional multiple hypotheses testing. We propose a false discovery rate specific to the problem under consideration, and construct a Bayes rule satisfying a false discovery rate criterion. The proposed method is compared with a traditional rule through simulation studies. We apply our methodology to two real examples involving microRNAs; where in one example the targeted genes are simultaneously downregulated under both experiments, and in the other the targeted genes are downregulated in one experiment and upregulated in the other experiment. We also discuss how the proposed methodology can be extended to more than two experiments.

Testing multiple hypotheses with skewed alternatives

In many practical cases of multiple hypothesis problems, it can be expected that the alternatives are not symmetrically distributed. If it is known a priori that the distributions of the alternatives are skewed, we show that this information yields high power procedures as compared to the procedures based on symmetric alternatives when testing multiple hypotheses. We propose a Bayesian decision theoretic rule for multiple directional hypothesis testing, when the alternatives are distributed as skewed, under a constraint on a mixed directional false discovery rate. We compare the proposed rule with a frequentist's rule of Benjamini and Yekutieli (2005) using simulations. We apply our method to a well-studied HIV dataset.

Bayesian analysis of two diagnostic methods for paediatric ringworm infections in a teaching hospital

Quantitatively, conventional methods of diagnosis of tinea capitis or paediatric ringworm, microscopic and culture tests were evaluated with Bayes rule. This analysis would help in quantifying the pervasive errors in each diagnostic method, particularly the microscopic method, as a long-term treatment would be involved to eradicate the infection by the use of a particular antifungal chemotherapy. Secondly, the analysis of clinical data would help in obtaining digitally the fallible standard of the microscopic test method, as the culture test method is taken as gold standard. Test results of 51 paediatric patients were of 4 categories: 21 samples were true positive (both tests positive), and 13 were true negative; the rest samples comprised both 14 false positive (microscopic test positivity with culture test negativity) and 3 false negative (microscopic test negativity with culture test positivity) samples. The prevalence of tinea infection was 47.01% in the population of 51 children. The microscopic test of a sample was efficient by 87.5%, in arriving at a positive result on diagnosis, when its culture test was positive; and, this test was efficient by 76.4%, in arriving at a negative result, when its culture test was negative. But, the post-test probability value of a sample with both microscopic and culture tests would be correct in distinguishing a sample from a sick or a healthy child with a chance of 71.5%. However, since the sensitivity of the analysis is 87.5%, the microscopic test positivity would be easier to detect in the presence of infection. In conclusion, it could be stated that Trychophyton rubrum was the most prevalent species; sensitivity and specificity of treating the infection, by antifungal therapy before ascertaining by the culture method remain as 0.8751 and 0.7642, respectively. A correct/coveted diagnostic method of fungal infection would be could be achieved by modern molecular methods (matrix-assisted laser desorption ionisation-time of flight mass spectrometry or fluorescence in situ hybridization or enzyme-linked immunosorbent assay [ELISA] or restriction fragment length polymorphism or DNA/RNA probes of known fungal taxa) in advanced laboratories.

Variable Selection in Nonparametric Classification via Measurement Error Model Selection Likelihoods

Using the relationships among ridge regression, LASSO estimation, and measurement error attenuation as motivation, a new measurement-error-model-based approach to variable selection is developed. After describing the approach in the familiar context of linear regression, we apply it to the problem of variable selection in nonparametric classification, resulting in a new kernel-based classifier with LASSO-like shrinkage and variable-selection properties. Finite-sample performance of the new classification method is studied via simulation and real data examples, and consistency of the method is studied theoretically. Supplementary materials for the paper are available online.

Statistical Evaluation of Two Microbiological Diagnostic Methods of Pulmonary Tuberculosis After Implementation of a Directly Observed Treatment Short-course Program

OBJECTIVES

To evaluate the diagnostic accuracy of smear and culture tests of clinical samples of pulmonary tuberculosis after the introduction of the directly observed treatment short-course (DOTS) program.

METHODS

Using sputum samples from 572 individuals as a self-selected population, both Ziehl-Neelsen staining and culturing on Lowenstein-Jensen medium were carried out as diagnostic procedures. Using Bayes' rule, the obtained data set was analyzed.

RESULTS

Of the 572 samples, 33 (0.05769) were true positive (results of both tests positive) cases; 22 samples (0.03846) were false positive (smear test positive and culture test negative) cases; 62 samples (0.10839) were false negative (smear test negative and culture test positive) cases; and 455 samples (0.79545) were true negative (results of both tests negative) cases. Values of test statistics, sensitivity, and specificity were used to compute several inherent other Bayesian test statistics. The a priori probability or prevalence value of tuberculosis in the targeted population was 0.166. The a posteriori probability value computed arithmetically was 0.6614 and that obtained by the graphical method was 0.62.

CONCLUSIONS

The smear test was found to be dependable for 95.4% with stable TB infections, and it was not dependable for 34.7% without stable TB infections. The culture test could be regarded as the gold standard for 96.15% as seen with the data set, which was obtained after the implementation of the DOTS program.