RESUMO
OBJECTIVES: : Although numerous risk prediction models have been proposed, few such models have been developed using neural network-based survival analysis. We developed risk prediction models for three cardiovascular disease risk factors (diabetes mellitus, hypertension, and dyslipidemia) among a working-age population in Japan using DeepSurv, a deep feed-forward neural network. METHODS: : Data were obtained from the Japan Epidemiology Collaboration on Occupational Health Study. A total of 51â258, 44â197, and 31â452 individuals were included in the development of risk models for diabetes mellitus, hypertension, and dyslipidemia, respectively; two-thirds of whom were used to develop prediction models, and the rest were used to validate the models. We compared the performances of DeepSurv-based models with those of prediction models based on the Cox proportional hazards model. RESULTS: : The area under the receiver-operating characteristic curve was 0.878 [95% confidence interval (CI)â=â0.864-0.892] for diabetes mellitus, 0.835 (95% CIâ=â0.826-0.845) for hypertension, and 0.826 (95% CIâ=â0.817-0.835) for dyslipidemia. Compared with the Cox proportional hazards-based models, the DeepSurv-based models had better reclassification performance [diabetes mellitus: net reclassification improvement (NRI)â=â0.474, P â≤â0.001; hypertension: NRIâ=â0.194, P â≤â0.001; dyslipidemia: NRIâ=â0.397, P â≤â0.001] and discrimination performance [diabetes mellitus: integrated discrimination improvement (IDI)â=â0.013, P â≤â0.001; hypertension: IDIâ=â0.007, P â≤â0.001; and dyslipidemia: IDIâ=â0.043, P â≤â0.001]. CONCLUSION: : This study suggests that DeepSurv has the potential to improve the performance of risk prediction models for cardiovascular disease risk factors.
Assuntos
Doenças Cardiovasculares , Diabetes Mellitus , Dislipidemias , Hipertensão , Humanos , Fatores de Risco , Doenças Cardiovasculares/etiologia , Diabetes Mellitus/epidemiologia , Hipertensão/complicações , Redes Neurais de Computação , Dislipidemias/complicaçõesRESUMO
Several algorithms have been proposed for modeling a gene regulatory network from a time-series expression dataset, but these have been used in relatively few studies because experimental cost often restricts the number of sampling time points to less than that of genes by more than one order of magnitude. In order to reduce the number of parameters for network modeling, we propose a method for grouping genes by both temporal expression pattern and biological function, modeling interactions between the gene groups by a dynamic Bayesian network approach. Results from applying the method to a gene expression dataset on human organogenesis demonstrate that more biologically plausible results can be obtained by modeling an interaction network for groups of genes than by modeling that for single genes.
Assuntos
Algoritmos , Embrião de Mamíferos/metabolismo , Regulação da Expressão Gênica no Desenvolvimento , Redes Reguladoras de Genes , Genes , Organogênese , Teorema de Bayes , Embrião de Mamíferos/citologia , Perfilação da Expressão Gênica , Humanos , Modelos Estatísticos , Fatores de TempoRESUMO
Metacontrast is a visual illusion in which the visibility of a target stimulus is virtually lost when immediately followed by a nonoverlapping mask stimulus. For a colored target, metacontrast is color-selective, with target visibility markedly reduced when the mask and target are the same color, but only slightly reduced when the colors differ. This study investigated neural correlates of color-selective metacontrast for cone-opponent red and green stimuli in the human V1, V2, and V3 using functional magnetic resonance imaging. Neural activity was suppressed when the target was rendered less visible by the same-colored mask, and the suppression was localized in the cortical region retinotopically representing the target, correlating with the perceptual topography of visibility/invisibility rather than the physical topography of the stimulus. Retinotopy-based group analysis found that activity suppression was statistically significant for V2 and V3 and that its localization to the target region was statistically significant for V2. These results suggest that retinotopic color representations in early visual areas, especially in V2, are closely linked to the visibility of color.