Evaluation of a decided sample size in machine learning applications.

BACKGROUND: An appropriate sample size is essential for obtaining a precise and reliable outcome of a study. In machine learning (ML), studies with inadequate samples suffer from overfitting of data and have a lower probability of producing true effects, while the increment in sample size increases the accuracy of prediction but may not cause a significant change after a certain sample size. Existing statistical approaches using standardized mean difference, effect size, and statistical power for determining sample size are potentially biased due to miscalculations or lack of experimental details. This study aims to design criteria for evaluating sample size in ML studies. We examined the average and grand effect sizes and the performance of five ML methods using simulated datasets and three real datasets to derive the criteria for sample size. We systematically increase the sample size, starting from 16, by randomly sampling and examine the impact of sample size on classifiers' performance and both effect sizes. Tenfold cross-validation was used to quantify the accuracy. RESULTS: The results demonstrate that the effect sizes and the classification accuracies increase while the variances in effect sizes shrink with the increment of samples when the datasets have a good discriminative power between two classes. By contrast, indeterminate datasets had poor effect sizes and classification accuracies, which did not improve by increasing sample size in both simulated and real datasets. A good dataset exhibited a significant difference in average and grand effect sizes. We derived two criteria based on the above findings to assess a decided sample size by combining the effect size and the ML accuracy. The sample size is considered suitable when it has appropriate effect sizes (≥ 0.5) and ML accuracy (≥ 80%). After an appropriate sample size, the increment in samples will not benefit as it will not significantly change the effect size and accuracy, thereby resulting in a good cost-benefit ratio. CONCLUSION: We believe that these practical criteria can be used as a reference for both the authors and editors to evaluate whether the selected sample size is adequate for a study.

Low-frequency oscillations in cortical level to help diagnose task-specific dystonia.

Task-specific dystonia is a neurological movement disorder that abnormal contractions of muscles result in the twisting of fixed postures or muscle spasm during specific tasks. Due to the rareness and the pathophysiology of the disease, there is no test to confirm the diagnosis of task-specific dystonia, except comprehensive observations by the experts. Evidence from neural electrophysiological data suggests that enhanced low frequency (4-12 Hz) oscillations in the subcortical structure of the globus pallidus were associated with the pathological abnormalities concerning ß and γ rhythms in motor areas and motor cortical network in patients with task-specific dystonia. However, whether patients with task-specific dystonia have any low-frequency abnormalities in motor cortical areas remains unclear. In this study, we hypothesized that low-frequency abnormalities are present in core motor areas and motor cortical networks in patients with task-specific dystonia during performing the non-symptomatic movements and those low-frequency abnormalities can help the diagnosis of this disease. We tested this hypothesis by using EEG, effective connectivity analysis, and a machine learning method. Fifteen patients with task-specific dystonia and 15 healthy controls were recruited. The machine learning method identified 8 aberrant movement-related network connections concerning low frequency, ß and γ frequencies, which enabled the separation of the data of patients from those of controls with an accuracy of 90%. Importantly, 7 of the 8 aberrant connections engaged the premotor area contralateral to the affected hand, suggesting an important role of the premotor area in the pathological abnormities. The patients exhibited significantly lower low frequency activities during the movement preparation and significantly lower ß rhythms during movements compared with healthy controls in the core motor areas. Our findings of low frequency- and ß-related abnormalities at the cortical level and aberrant motor network could help diagnose task-specific dystonia in the clinical setting, and the importance of the contralesional premotor area suggests its diagnostic potential for task-specific dystonia.

High-Availability Computing Platform with Sensor Fault Resilience.

Modern computing platforms usually use multiple sensors to report system information. In order to achieve high availability (HA) for the platform, the sensors can be used to efficiently detect system faults that make a cloud service not live. However, a sensor may fail and disable HA protection. In this case, human intervention is needed, either to change the original fault model or to fix the sensor fault. Therefore, this study proposes an HA mechanism that can continuously provide HA to a cloud system based on dynamic fault model reconstruction. We have implemented the proposed HA mechanism on a four-layer OpenStack cloud system and tested the performance of the proposed mechanism for all possible sets of sensor faults. For each fault model, we inject possible system faults and measure the average fault detection time. The experimental result shows that the proposed mechanism can accurately detect and recover an injected system fault with disabled sensors. In addition, the system fault detection time increases as the number of sensor faults increases, until the HA mechanism is degraded to a one-system-fault model, which is the worst case as the system layer heartbeating.

ß-1,4-Galactosyltransferase III suppresses ß1 integrin-mediated invasive phenotypes and negatively correlates with metastasis in colorectal cancer.

Metastasis often occurs in colorectal cancer (CRC) patients and is the main difficulty in cancer treatment. The upregulation of poly-N-acetyllactosamine-related glycosylation is found in CRC patients and is associated with progression and metastasis in cancer. ß-1,4-Galactosyltransferase III (B4GALT3) is an enzyme responsible for poly-N-acetyllactosamine synthesis, and therefore, we investigated its expression in CRC patients. We found that B4GALT3 negatively correlated with poorly differentiated histology (P < 0.001), advanced stages (P = 0.0052), regional lymph node metastasis (P = 0.0018) and distant metastasis (P = 0.0463) in CRC patients. B4GALT3 overexpression in CRC cells suppressed cell migration, invasion and adhesion, whereas B4GALT3 knockdown enhanced malignant cell phenotypes. The ß1 integrin-blocking antibody reversed the B4GALT3-mediated increase in cell invasion. B4GALT3 expression altered glycosylation on the N-glycan of ß1 integrin probably through changes in poly-N-acetyllactosamine expression. Furthermore, more activated ß1 integrin along with the activation of its downstream signaling transduction were found in B4GALT3 knockdown cells, whereas overexpression of B4GALT3 suppressed the expression of active ß1 integrin and inhibited its downstream signaling. Our results suggest that B4GALT3 is negatively associated with CRC metastasis and suppresses cell invasiveness through inhibiting activation of ß1 integrin.

The optimal dose of oral tranexamic acid in melasma: A network meta-analysis.

BACKGROUND: Melasma is a chronic skin condition that adversely impacts quality of life. Although many therapeutic modalities are available there is no single best treatment for melasma. Oral tranexamic acid has been used for the treatment of this condition but its optimal dose is yet to be established. OBJECTIVES: We used network meta-analysis to determine the optimal dose of oral tranexamic acid for the treatment of melasma. METHODS: We conducted a comprehensive search of all studies of oral tranexamic acid for the treatment of melasma up to September 2020 using PubMed, EMBASE and the Cochrane Library database. The quality of the studies was evaluated using the Jadad score and the Cochrane's risk of bias assessment tool. Only high quality randomised controlled trials were selected. Some studies lacked standard deviation of changes from baseline and these were estimated using the correlation coefficient obtained from another similar study. RESULTS: A total of 92 studies were identified of which 6 randomized controlled trials comprising 599 patients were included to form 3 pair-wise network comparisons. The mean age of the patients in these studies ranged from 30.3 to 46.5 years and the treatment duration ranged from 8 to 12 weeks. The Jadad scores ranged from 5 to 8. The optimal dose and duration of oral tranexamic acid was estimated to be 750 mg per day for 12 consecutive weeks. LIMITATIONS: Some confounding factors might not have been described in the original studies. Although clear rules were followed, the Melasma Area and Severity Index and the modified Melasma Area and Severity Index were scored by independent physicians and hence inter-observer bias could not be excluded. CONCLUSION: Oral tranexamic acid is a promising drug for the treatment of melasma. This is the first network meta-analysis to determine the optimal dose of this drug and to report the effects of different dosages. The optimal dose is 250 mg three times per day for 12 weeks, but 250 mg twice daily may be an acceptable option in poorly adherent patients. Our findings will allow physicians to balance drug effects and medication adherence. Personalized treatment plans are warranted.

A dynamic causal model for evoked and induced responses.

Neuronal responses exhibit two stimulus or task-related components: evoked and induced. The functional role of induced responses has been ascribed to 'top-down' modulation through backward connections and lateral interactions; as opposed to the bottom-up driving processes that may predominate in evoked components. The implication is that evoked and induced components may reflect different neuronal processes. The conventional way of separating evoked and induced responses assumes that they can be decomposed linearly; in that induced responses are the average of the power minus the power of the average (the evoked component). However, this decomposition may not hold if both components are generated by nonlinear processes. In this work, we propose a Dynamic Causal Model that models evoked and induced responses at the same time. This allows us to explain both components in terms of shared mechanisms (coupling) and changes in coupling that are necessary to explain any induced components. To establish the face validity of our approach, we used Bayesian Model Selection to show that the scheme can disambiguate between models of synthetic data that did and did not contain induced components. We then repeated the analysis using MEG data during a hand grip task to ask whether induced responses in motor control circuits are mediated by 'top-down' or backward connections. Our result provides empirical evidence that induced responses are more likely to reflect backward message passing in the brain, while evoked and induced components share certain characteristics and mechanisms.

Distinguishing the Visual Working Memory Training and Practice Effects by the Effective Connectivity During n-back Tasks: A DCM of ERP Study.

Visual working memory (WM) training and practice can result in improved task performance and increased P300 amplitude; however, only training can yield N160 enhancements. N160 amplitudes are related to the spatial attention, the detection of novelty and the inhibitory control, while P300 amplitudes are related to the selective attention. Therefore, it could be speculated that the mechanisms underlying N160 and P300 production may differ to accommodate to their functions. Based on the different N160 engagements and different functional roles of N160 and P300, we hypothesized that the effects of visual WM training and practice can be dissociated by their brain effective connectivity patterns. We compared different neural connectivity configurations for the main task-related brain activities including N160 and P300 during the visual three-back task in subjects after visual WM training (the WM group) and after repetitive task practice (the control group). The behavioral result shows significantly greater improvement in accuracy after training and suggests that visual WM training can boost the learning process of this simple task. The N160 peak amplitude increased significantly after training over the anterior and posterior brain areas but decreased after practice over the posterior areas, indicating different mechanisms for mediating the training and practice effects. In support of our hypothesis, we observed that visual WM training alters the frontal-parietal connections, which comprise the executive control network (ECN) and the dorsal attention network (DAN), whereas practice modulates the parietal-frontal connections underpinning P300 production for selective attention. It should be noted that the analytic results in this study are conditional on the plausible models being tested and the experimental settings. Studies that employ different tasks, devices and plausible models may lead to different results. Nevertheless, our findings provide a reference for distinguishing the visual WM training and practice effects by the underlying neuroplasticity.

Bone health among older adults in Taiwan.

BACKGROUND AND PURPOSE: There has been much discussion about the risk factors for osteoporosis, but studies involving elderly population in Taiwan are minimal. We aimed to describe variables related to osteoporosis among community dwelling older people in Taiwan. METHODS: This is a cross-sectional study. The 671 participants were randomly selected from 3680 examinees of the annual Senior Citizens Health Examination in year 2010. Participants were interviewed with a detailed questionnaire, and 91 of them were invited for dual-energy X-ray absorptiometry (DXA). Predictor variables included age, gender and clinical risk factors for osteoporosis. The main outcome was osteoporosis confirmed by DXA. RESULTS: The mean age of the participants was 75.7±6.4years old. Overall, the most prevalent variables for osteoporosis were height loss in adulthood (41.0%), lack of dairy products or calcium supplements (32.0%) and insufficient physical activity (10.4%). In multivariate models, we found that underweight (OR=9.80) and lack of dairy products/calcium supplements (OR=3.68) were the main variables for osteoporosis. In the subgroup analysis involving only women, underweight (OR=14.60) was the main variable. DISCUSSION: Among community-dwelling older people in Taiwan, osteoporosis was mainly associated with underweight and lack of dairy products or calcium supplements. CONCLUSION: We suggest using the key questions of underweight and dietary pattern in clinical settings to identify high risk people who are candidates for further BMD exam.

EEG-based motor network biomarkers for identifying target patients with stroke for upper limb rehabilitation and its construct validity.

Rehabilitation is the main therapeutic approach for reducing poststroke functional deficits in the affected upper limb; however, significant between-patient variability in rehabilitation efficacy indicates the need to target patients who are likely to have clinically significant improvement after treatment. Many studies have determined robust predictors of recovery and treatment gains and yielded many great results using linear approachs. Evidence has emerged that the nonlinearity is a crucial aspect to study the inter-areal communication in human brains and abnormality of oscillatory activities in the motor system is linked to the pathological states. In this study, we hypothesized that combinations of linear and nonlinear (cross-frequency) network connectivity parameters are favourable biomarkers for stratifying patients for upper limb rehabilitation with increased accuracy. We identified the biomarkers by using 37 prerehabilitation electroencephalogram (EEG) datasets during a movement task through effective connectivity and logistic regression analyses. The predictive power of these biomarkers was then tested by using 16 independent datasets (i.e. construct validation). In addition, 14 right handed healthy subjects were also enrolled for comparisons. The result shows that the beta plus gamma or theta network features provided the best classification accuracy of 92%. The predictive value and the sensitivity of these biomarkers were 81.3% and 90.9%, respectively. Subcortical lesion, the time poststroke and initial Wolf Motor Function Test (WMFT) score were identified as the most significant clinical variables affecting the classification accuracy of this predictive model. Moreover, 12 of 14 normal controls were classified as having favourable recovery. In conclusion, EEG-based linear and nonlinear motor network biomarkers are robust and can help clinical decision making.

A novel approach for discovering condition-specific correlations of gene expressions within biological pathways by using cloud computing technology.

Microarrays are widely used to assess gene expressions. Most microarray studies focus primarily on identifying differential gene expressions between conditions (e.g., cancer versus normal cells), for discovering the major factors that cause diseases. Because previous studies have not identified the correlations of differential gene expression between conditions, crucial but abnormal regulations that cause diseases might have been disregarded. This paper proposes an approach for discovering the condition-specific correlations of gene expressions within biological pathways. Because analyzing gene expression correlations is time consuming, an Apache Hadoop cloud computing platform was implemented. Three microarray data sets of breast cancer were collected from the Gene Expression Omnibus, and pathway information from the Kyoto Encyclopedia of Genes and Genomes was applied for discovering meaningful biological correlations. The results showed that adopting the Hadoop platform considerably decreased the computation time. Several correlations of differential gene expressions were discovered between the relapse and nonrelapse breast cancer samples, and most of them were involved in cancer regulation and cancer-related pathways. The results showed that breast cancer recurrence might be highly associated with the abnormal regulations of these gene pairs, rather than with their individual expression levels. The proposed method was computationally efficient and reliable, and stable results were obtained when different data sets were used. The proposed method is effective in identifying meaningful biological regulation patterns between conditions.

Accelerating Computation of DCM for ERP in MATLAB by External Function Calls to the GPU.

This study aims to improve the performance of Dynamic Causal Modelling for Event Related Potentials (DCM for ERP) in MATLAB by using external function calls to a graphics processing unit (GPU). DCM for ERP is an advanced method for studying neuronal effective connectivity. DCM utilizes an iterative procedure, the expectation maximization (EM) algorithm, to find the optimal parameters given a set of observations and the underlying probability model. As the EM algorithm is computationally demanding and the analysis faces possible combinatorial explosion of models to be tested, we propose a parallel computing scheme using the GPU to achieve a fast estimation of DCM for ERP. The computation of DCM for ERP is dynamically partitioned and distributed to threads for parallel processing, according to the DCM model complexity and the hardware constraints. The performance efficiency of this hardware-dependent thread arrangement strategy was evaluated using the synthetic data. The experimental data were used to validate the accuracy of the proposed computing scheme and quantify the time saving in practice. The simulation results show that the proposed scheme can accelerate the computation by a factor of 155 for the parallel part. For experimental data, the speedup factor is about 7 per model on average, depending on the model complexity and the data. This GPU-based implementation of DCM for ERP gives qualitatively the same results as the original MATLAB implementation does at the group level analysis. In conclusion, we believe that the proposed GPU-based implementation is very useful for users as a fast screen tool to select the most likely model and may provide implementation guidance for possible future clinical applications such as online diagnosis.

ß-1,4-Galactosyltransferase III enhances invasive phenotypes via ß1-integrin and predicts poor prognosis in neuroblastoma.

PURPOSE: Neuroblastoma (NB) is a neural crest-derived tumor that commonly occurs in childhood. ß-1,4-Galactosyltransferase III (B4GALT3) is highly expressed in human fetal brain and is responsible for the generation of poly-N-acetyllactosamine, which plays a critical role in tumor progression. We therefore investigated the expression and role of B4GALT3 in NB. EXPERIMENTAL DESIGN: We examined B4GALT3 expression in tumor specimens from 101 NB patients by immunohistochemistry and analyzed the correlation between B4GALT3 expression and clinicopathologic factors or survival. The functional role of B4GALT3 expression was investigated by overexpression or knockdown of B4GALT3 in NB cells for in vitro and in vivo studies. RESULTS: We found that B4GALT3 expression correlated with advanced clinical stages (P = 0.040), unfavorable Shimada histology (P < 0.001), and lower survival rate (P < 0.001). Multivariate analysis showed that B4GALT3 expression is an independent prognostic factor for poor survival of NB patients. B4GALT3 overexpression increased migration, invasion, and tumor growth of NB cells, whereas B4GALT3 knockdown suppressed the malignant phenotypes of NB cells. Mechanistic investigation showed that B4GALT3-enhanced migration and invasion were significantly suppressed by ß1-integrin blocking antibody. Furthermore, B4GALT3 overexpression increased lactosamine glycans on ß1-integrin, increased expression of mature ß1-integrin via delayed degradation, and enhanced phosphorylation of focal adhesion kinase. Conversely, these properties were decreased by knockdown of B4GALT3 in NB cells. CONCLUSIONS: Our findings suggest that B4GALT3 predicts an unfavorable prognosis for NB and may regulate invasive phenotypes through modulating glycosylation, degradation, and signaling of ß1-integrin in NB cells.