Bayesian Network Learning Approach Research Articles

Relationship between built environment characteristics of TOD and subway ridership: A causal inference and regression analysis of the Beijing subway

Numerous studies suggest that built environments have impacts on transit ridership, but few consider the causal connection between them. The goal of this study is to examine the causal relationship between the built environment and subway ridership and then to re-examine the impacts. In the case of the Beijing subway system, we use the Bayesian Network learning approach to examine the causal relationship between built environment characteristics and ridership. Based on the causality analysis, we further explore the impact of the built environment on subway ridership using Ordinary Least Squares and Geographically Weighted Regression models. Findings reveal the causal impact of employment density and public transport accessibility on alighting ridership during the morning peak. The result of the correlation analysis between the morning-peak alighting ridership and other variables shows that higher employment density and public transport accessibility produce more travel demand. The regression model also indicates that the effects of the built environment on ridership vary across space. Last but not least, the results of the model performance tests show that the model constructed from the indicators obtained from the causality screening is reliable.

PEWOBS: An efficient Bayesian network learning approach based on permutation and extensible ordering-based search

How to efficiently construct Bayesian network (BN) is a major research hotspot in the field of artificial intelligence. Ordering-based search (OBS) algorithm is a kind of search scoring-based BN structure learning algorithm with good performance. This kind of algorithm constructs the BN structure by optimizing the ordering of variables iteratively. However, it has the problems of poor initial ordering quality, insufficient ordering optimization and low accuracy. In order to solve these problems, this paper proposes a Bayesian network structure learning approach (PEWOBS) based on full permutation and extensible ordering-based search. It firstly uses the local learning and pruning method to process the dataset with small and large number of variables respectively, and obtains the candidate parent sets of each variable efficiently. Then it obtains the high quality of initial ordering in consideration of the global benefits and individual losses. On the basis of the initial optimization of ordering by the extensible window operator, the second optimization of sub-ordering is performed using the full permutation operator. The final BN structure with the largest score will be obtained according to the final ordering. Experimental results show that PEWOBS can get a better network structure in the first iteration. It has better network learning accuracy and efficiency compared with several kinds of existing OBS approaches.Availability and implementation: codes and experiment dataset are available at: http://122.205.95.139/PEWOBS/.

Brain connectivity during simulated balance in older adults with and without Parkinson's disease

Individuals with Parkinson’s disease often experience postural instability, a debilitating and largely treatment-resistant symptom. A better understanding of the neural substrates contributing to postural instability could lead to more effective treatments. Constraints of current functional neuroimaging techniques, such as the horizontal orientation of most MRI scanners (forcing participants to lie supine), complicates investigating cortical and subcortical activation patterns and connectivity networks involved in healthy and parkinsonian balance control. In this cross-sectional study, we utilized a newly-validated MRI-compatible balance simulator (based on an inverted pendulum) that enabled participants to perform balance-relevant tasks while supine in the scanner. We utilized functional MRI to explore effective connectivity underlying static and dynamic balance control in healthy older adults (n = 17) and individuals with Parkinson’s disease while on medication (n = 17). Participants performed four tasks within the scanner with eyes closed: resting, proprioceptive tracking of passive ankle movement, static balancing of the simulator, and dynamic responses to random perturbations of the simulator. All analyses were done in the participant's native space without spatial transformation to a common template. Effective connectivity between 57 regions of interest was computed using a Bayesian Network learning approach with false discovery rate set to 5%. The first 12 principal components of the connection weights, binomial logistic regression, and cross-validation were used to create 4 separate models: contrasting static balancing vs {rest, proprioception} and dynamic balancing vs {rest, proprioception} for both controls and individuals with Parkinson’s disease. In order to directly compare relevant connections between controls and individuals with Parkinson’s disease, we used connections relevant for predicting a task in either controls or individuals with Parkinson’s disease in logistic regression with Least Absolute Shrinkage and Selection Operator regularization. During dynamic balancing, we observed decreased connectivity between different motor areas and increased connectivity from the brainstem to several cortical and subcortical areas in controls, while individuals with Parkinson’s disease showed increased connectivity associated with motor and parietal areas, and decreased connectivity from brainstem to other subcortical areas. No significant models were found for static balancing in either group. Our results support the notion that dynamic balance control in individuals with Parkinson’s disease relies more on cortical motor areas compared to healthy older adults, who show a preference of subcortical control during dynamic balancing.

Predicting building damages in mega-disasters under uncertainty: An improved Bayesian network learning approach

This paper develops a hybrid approach that integrates the cloud model and Bayesian networks (BNs) to predict building damages induced by earthquakes under uncertainty. The cloud model is used to perform data discretization for modeling information losses and uncertainties. The learned BN from the historical data by using the Peter-Clark (PC) algorithm is validated by using the k-fold cross-validation. Three types of analysis, including predictive, diagnosis, and sensitivity analysis, are conducted for new knowledge discovery. A selected dataset consisting of 9920 buildings in the 2015 Nepal earthquake is used to demonstrate the adaptability and significance of the proposed approach. Results imply that (1) The constructed BN displays a high precision in predicting earthquake-induced building damage, where the accuracy of the 10-fold cross-validation ranges from 0.89 to 0.92. (2) Building structures and foundation materials are identified as critical factors leading to severe building damages. (3) The proposed approach possesses better prediction capabilities than the conventional BN approach, where an average improvement of 3.5 % in terms of the average accuracy is achieved. This research provides a data-driven perspective to perceive the magnitude of building damages subjected to uncertainties and complexities.

Rapidly Learning Bayesian Networks for Complex System Diagnosis: A Reinforcement Learning Directed Greedy Search Approach

Bayesian networks are a popular diagnosis method, whose structures are usually defined by human experts and parameters are learned from data. For the increasing complexity of modern systems, building their structures based on physical behaviors is becoming a difficult task. However, the improvement of data collection techniques motivates learning their structures from data, where greedy search is a typical iterative method. In each iteration, it generates multiple structure candidates by modifying one edge, evaluates these structures by scores based on data and selects the best structure for the next iteration. This method is costly because there are too many structures to be evaluated. To solve this problem, we frame the traditional greedy search by Markov decision process and propose an efficient Bayesian network learning approach by integrating reinforcement learning into it. In our approach, a convolutional neural network is employed as the value function to approximate scores. Before evaluating structures using data, the neural network is used to predict scores. Structure candidates with a low predicted score are discarded. By avoiding unnecessary computation, the cooperation of reinforcement learning and greedy search effectively improves the learning efficiency. Two systems, a 10-tank system with 21 monitored variables and the classic Tennessee Eastman process with 52 variables, are employed to demonstrate our approach. The experiment results indicate that the computation cost of our method was reduced by 30%~50%, and the diagnosis accuracy was almost the same.

Abstract 4294: One-carbon metabolism and colorectal cancer risk according to molecular subtypes: a Bayesian network learning approach

Abstract Background: Colorectal cancer (CRC) is a heterogeneous disease caused by a number of distinct pathways defined by genetic and epigenetic events. These molecular subtypes differ in prognosis and treatment options, and recent studies also imply differences in etiology. One-carbon metabolism has been extensively studied in relation to incident CRC risk because of its involvement in nucleotide synthesis and methylation reactions, yet only a few studies have evaluated if associations differ over CRC subtypes. In this study, we investigated a targeted panel of circulating metabolites and a set of single nucleotide polymorphisms involved in one-carbon metabolism in relation to the risk of overall CRC and CRC subtypes approximated by mutations in the BRAF and KRAS oncogenes, microsatellite instability (MSI), or CpG island methylator phenotype (CIMP). Methods: This was a prospective population-based nested case control study of 613 CRC cases and 1190 matched controls. The data included plasma concentrations of 14 one-carbon metabolites, 17 related single nucleotide polymorphisms, and other lifestyle-related information for all subjects, and clinical and molecular information for the tumors. In order to account for interactions and dependencies between variables, we applied multivariate Bayesian network learning in combination with more traditional univariate multinomial logistic regression to investigate independent relations to subtype specific CRC risk. Results: Preliminary findings show that vitamin B6 (Pyridoxal 5-phosphate, PLP) and vitamin B2 (riboflavin) had the strongest independent relations to overall CRC risk in the multivariate analyses. Folate, glycine, vitamin B6, and the RFC1 80G>A polymorphism displayed signs of independent relations to CRC subtypes. In univariate multinomial logistic regression models, plasma folate expressed the strongest etiologic heterogeneity, and was more clearly related to the risk of CIMP low/high tumors, tumors with MSI, or tumors without KRAS mutation, respectively. Conclusions: This study investigated a large set of prediagnostic circulating metabolites and single nucleotide polymorphisms related to one-carbon metabolism using multivariate statistical learning. We found that some components showed signs of subtype specific relations, of which the association for folate was clearest. Our results suggest that the role of one-carbon metabolism in colorectal carcinogenesis may differ between specific molecular pathways. Citation Format: Robin Myte, Björn Gylling, Jenny Häggström, Jörn Schneede, Per Magne Ueland, Göran Hallmans, Ingegerd Johansson, Richard Palmqvist, Bethany Van Guelpen. One-carbon metabolism and colorectal cancer risk according to molecular subtypes: a Bayesian network learning approach. [abstract]. In: Proceedings of the 107th Annual Meeting of the American Association for Cancer Research; 2016 Apr 16-20; New Orleans, LA. Philadelphia (PA): AACR; Cancer Res 2016;76(14 Suppl):Abstract nr 4294.

Abstract 1592: Identifying novel drivers of the epithelial-to-mesenchymal transition across multiple cancer types: from bioinformatics to the bench

Abstract To emigrate from a primary tumor to a distant site, a proliferating cancer cell must acquire the phenotypic traits necessary for migration and invasion. This shift from “growing” to “going,” termed epithelial-to-mesenchymal transition (EMT), may be a genetic or epigenetic phenomenon, ultimately characterized by altered cellular function that promotes metastasis. In addition to its role in malignant progression, EMT is integral in development and for healthy wound healing. EMT has been described in all carcinomas, and, because it of its established physiologic role, we hypothesize that there are a set of central EMT regulators that are universal to all carcinoma types. To identify global regulators of EMT, we integrated data from 15 published gene expression microarray studies that include a total of 49 epithelial and 46 mesenchymal cell line samples across 6 malignant tissue types (breast, prostate, colon, esophageal, liver, retinal pigment) and with various in vitro EMT induction strategies. While accounting for batch effects and other technical variability inherent in gene expression data, we performed differential expression analysis to identify genes that vary significantly between epithelial and mesenchymal states. Importantly, we found differential expression of established EMT markers (e.g. CDH1, ZEB1) validating our approach. We also identified genes that had not previously been implicated in cancer progression, representing novel candidate drivers of EMT, for example SARG (C1orf116). We found that SARG expression is decreased in high-grade cancer and metastatic disease, and is negatively associated with chemotherapy resistance across multiple cancer types (Oncomine). In an in vitro model of prostate cancer EMT, we found that SARG had >5-fold increased RNA and protein expression in PC3-epithelial cells. Knockdown of SARG expression in PC3-epithelial cells resulted in decreased gene expression of the epithelial-marker CDH1 and elevated expression of the mesenchymal marker CDH2, suggesting a role as a driver of the epithelial phenotype. In parallel with functional in vitro experiments, we have applied a Bayesian network learning approach to identify differentially expressed genes that are likely to play regulatory roles in EMT with causal influence on other genes. Importantly, this analysis provides valuable information regarding novel downstream effectors of known key regulators of EMT such as CDH1 and ELF3. Moreover, we also identified new EMT regulators such as CEP170 that have not been previously described. Experiments are underway to further explore the functional implications of these gene networks. This integrative approach of global gene expression analysis, network learning, and functional validation may identify causal genes and regulatory interactions in EMT, both expanding upon current knowledge and identifying novel drivers of this key metastatic process. Citation Format: Sarah R. Amend, Princy Parsana, James Hernandez, Alexis Battle, Kenneth J. Pienta. Identifying novel drivers of the epithelial-to-mesenchymal transition across multiple cancer types: from bioinformatics to the bench. [abstract]. In: Proceedings of the 107th Annual Meeting of the American Association for Cancer Research; 2016 Apr 16-20; New Orleans, LA. Philadelphia (PA): AACR; Cancer Res 2016;76(14 Suppl):Abstract nr 1592.

Leak Prediction Model for Water Distribution Networks Created Using a Bayesian Network Learning Approach

Water leakage in water distribution systems (WDSs) can bring various negative economic, environmental, and safety effects. Therefore, predicting water leakage is one of the most crucial tasks in water resource management; however, it is also one of the most challenging ones. Previous leakage-related studies have only focused on detecting existing leaks. This paper presents a novel model using expert structural expectation–maximisation, for predicting water leakage in WDSs. The model can take into account the uncertainty of leakage-related factors and balance the contribution of monitoring data and prior information in a Bayesian learning process to maximise leakage prediction accuracy. Moreover, the proposed method can indicate the most crucial factors affecting water leakage. The results of this study could benefit water utilities by aiding them in establishing an effective leakage control plan to minimise the risk of water leakage. A case study is presented to demonstrate the robustness and effectiveness of the proposed method.

CCGA-BN Constructor: A Bayesian Network Learning Approach

This paper presents a tool CCGA-BN Constructor for learning Bayesian network that uses cooperative coevolutionary genetic algorithm to learn Bayesian network structure from data. The problem has been broken down into two sub-problems: (a) to find the optimal nodes’ordering and (b) to find the optimal adjacency matrix of the graph. Both the sub-problems’ solutions are then combined to produce the optimal structure. CCGA-BN constructor used Bayesian score for networks having nodes with more than two states and BIC for network having bistate nodes. The findings of this paper are compared against the original structures and the results show a lot of promise.

Directionality of large-scale resting-state brain networks during eyes open and eyes closed conditions.

The present study examined directional connections in the brain among resting-state networks (RSNs) when the participant had their eyes open (EO) or had their eyes closed (EC). The resting-state fMRI data were collected from 20 healthy participants (9 males, 20.17 ± 2.74 years) under the EO and EC states. Independent component analysis (ICA) was applied to identify the separated RSNs (i.e., the primary/high-level visual, primary sensory-motor, ventral motor, salience/dorsal attention, and anterior/posterior default-mode networks), and the Gaussian Bayesian network (BN) learning approach was then used to explore the conditional dependencies among these RSNs. The network-to-network directional connections related to EO and EC were depicted, and a support vector machine (SVM) was further employed to identify the directional connection patterns that could effectively discriminate between the two states. The results indicated that the connections among RSNs are directionally connected within a BN during the EO and EC states. The directional connections from the salience network (SN) to the anterior/posterior default-mode networks and the high-level to primary-level visual network were the obvious characteristics of both the EO and EC resting-state BNs. Of the directional connections in BN, the directional connections of the salience and dorsal attention network (DAN) were observed to be discriminative between the EO and EC states. In particular, we noted that the properties of the salience and DANs were in opposite directions. Overall, the present study described the directional connections of RSNs using a BN learning approach during the EO and EC states, and the results suggested that the directionality of the attention systems (i.e., mainly for the salience and the DAN) in resting state might have important roles in switching between the EO and EC conditions.

Efficient Heuristics for Discriminative Structure Learning of Bayesian Network Classifiers

We introduce a simple order-based greedy heuristic for learning discriminative structure within generative Bayesian network classifiers. We propose two methods for establishing an order of N features. They are based on the conditional mutual information and classification rate (i.e., risk), respectively. Given an ordering, we can find a discriminative structure with O(Nk+1) score evaluations (where constant k is the tree-width of the sub-graph over the attributes). We present results on 25 data sets from the UCI repository, for phonetic classification using the TIMIT database, for a visual surface inspection task, and for two handwritten digit recognition tasks. We provide classification performance for both discriminative and generative parameter learning on both discriminatively and generatively structured networks. The discriminative structure found by our new procedures significantly outperforms generatively produced structures, and achieves a classification accuracy on par with the best discriminative (greedy) Bayesian network learning approach, but does so with a factor of ~10-40 speedup. We also show that the advantages of generative discriminatively structured Bayesian network classifiers still hold in the case of missing features, a case where generative classifiers have an advantage over discriminative classifiers.