Unbalanced Sample Sizes Research Articles

Readers' Comprehension of New Text Formats. Part I. Empirical Evidence from Russian and Research Methodology Behind It

In this article, we review a range of Russian-language studies investigating readers’ interaction with new text formats. We identify the key problems preventing proper generalization of such research data. These include: (1) the absence of agreed terminology for describing texts of new formats shared by most researchers, and (2) the lack of consistency in theoretical approaches for describing cognitive mechanisms underpinning processing of these types of texts by readers. The following factors further complicate comparison and systematization of the data obtained by different research groups: (1) the lack of explicitly described theoretical bases of the hypotheses formulated by researchers, (2) the lack of consistency in describing the putative connections between such theoretical backgrounds and empirically tested hypotheses, (3) heterogeneity of methodological approaches, (4) insufficient, unequal or unbalanced sample sizes, (5) low reliability of statistical approaches and tools used for data analysis, as well as (6) research artefacts confounding the results.

Technostress and time spent online. A cross-cultural comparison for teachers and students.

Research shows that ICT is beneficial for academics and students, aiding in overcoming distance barriers, streamlining administration, and improving teaching and learning processes. However, the negative impact of technology, particularly technostress, are garnering attention. In the context of the concerns about technostress among higher education institutions (HEI), the aim of the study is to analyze the technostress creators and inhibitors for university teachers and students in different European countries. The topical concept of digital well-being is seen as "a subjective individual experience of optimal balance between the benefits and drawbacks obtained from mobile connectivity, focusing on the personal perception of what amount of time spent using technology is optimal so that well-being is preserved". To explore specific aspects related to use of technology, two - parallel online surveys for academics (N = 446) and students (N = 660) from four European countries (Romania, Germany, Norway, and Finland) were conducted between November 2022 - January 2023. The surveys included the Technostress scale and the Technostress Inhibitors Scale, the Technology self-efficacy Scale, and a questionnaire focusing on socio-demographic aspects, work experience, academic field, dimensions related to the actual use of technology and participants perception on the optimal use of technology for work, learning or personal tasks, in terms of the period of day/week and amount of time spent. We also inquired about the social support given and received when using technology and the formal and informal rules, expectations, policies, punishments, and rewards regarding the use of technology. The findings suggest that the perceived optimal use of technology is significantly lower than the actual use for all the contexts. Overuse of technology was associated with technostress. Our results also showed that technology self-efficacy and social support from colleagues and teachers are negatively associated with technostress. Country differences regarding technostress and time spent online were also observed. Despite the needed caution in interpreting the results because of the unbalanced sample size across countries, the results could be used to develop research and support interventions within European countries to promote digital well-being, a better work-life balance with further positive effects on academic satisfaction and work/learning productivity.

Using shell shape analysis based on landmarks to trace the geographical origin of the common cockle (Cerastoderma edule)

Determining the geographical origin of seafood is crucial for regulators and fishing industries who seek to prevent commercial fraud, enforce food safety regulations, and encourage high standards in sustainable fisheries management. The cockle, Cerastoderma edule (Linnaeus 1768), is a key species in estuarine ecosystems and is harvested all over Europe. Therefore, traceability tools using quick and inexpensive techniques to identify the origin of this bivalve are of paramount importance to support law enforcement. In this work, we explore the potential of using Geometric Morphometric (GM) methods to identify the geographical origin of cockle specimens. This method is based on landmarks identified in the shell to trace the origin of specimens obtained in nearby aquatic systems (from <35 km to <250 km distance). Specimens were collected in five aquatic systems (Ria de Aveiro, the Tagus and Sado estuaries, and the Albufeira and Óbidos coastal lagoons) in Portugal. Shells were digitalized and 16 landmarks were identified in each right valve and analyzed using Generalized Procrustes Superimposition. The discriminating power for traceability of 12 statistical and machine learning methods was assessed based on the corresponding shape variables, using R and Python (Linear Discriminant Analysis (LDA), Canonical Variable Analysis (CVA), Principal Component Analysis (PCA), Between-Group PCA (bgPCA), Partial Least Squares Discriminant (PLSD), Classification Regression Tree (CRT), Logistic Regression (LR), Random Forest (RF), Gradient Boosting (GB), K Nearest Neighbors (KNN), Support Vector Machines (SVM), Extending Gradient Boosting (XGBoost) and Neural Networks (NNET). LDA, CVA, SVM, and NNET demonstrated greater accuracy and a F1-score >80%, even with a small and unbalanced sample size. The highest percentage of correctly assigned individuals was obtained in the Tagus estuary (mean 89%) and in the Albufeira lagoon (mean 93%), which were also the systems with more specimens measured (174 and 59 respectively), whereas the worst results were obtained in the Sado estuary (50%, 56 specimens). In the Albufeira coastal lagoon, the best classification methods reached 100% correct classifications. It further highlights the importance of establishing statistical standards, such as the ones developed in the current work, to evaluate different methods, as small changes in the procedure may cause substantial differences in the results and conclusions. The revision of previous works (presented as a table) showed often >90% of correct classification in both bivalves and gastropods, highlighting the potential of the techniques for other mollusks. Our results support the use of GM based on landmarks as a reliable tool for bivalve's traceability, since it is a quick, simple and inexpensive approach. Further research should extend these findings to other species and other shape analysis techniques.

Differential diagnosis of frontotemporal dementia subtypes with explainable deep learning on structural MRI.

Frontotemporal dementia (FTD) represents a collection of neurobehavioral and neurocognitive syndromes that are associated with a significant degree of clinical, pathological, and genetic heterogeneity. Such heterogeneity hinders the identification of effective biomarkers, preventing effective targeted recruitment of participants in clinical trials for developing potential interventions and treatments. In the present study, we aim to automatically differentiate patients with three clinical phenotypes of FTD, behavioral-variant FTD (bvFTD), semantic variant PPA (svPPA), and nonfluent variant PPA (nfvPPA), based on their structural MRI by training a deep neural network (DNN). Data from 277 FTD patients (173 bvFTD, 63 nfvPPA, and 41 svPPA) recruited from two multi-site neuroimaging datasets: the Frontotemporal Lobar Degeneration Neuroimaging Initiative and the ARTFL-LEFFTDS Longitudinal Frontotemporal Lobar Degeneration databases. Raw T1-weighted MRI data were preprocessed and parcellated into patch-based ROIs, with cortical thickness and volume features extracted and harmonized to control the confounding effects of sex, age, total intracranial volume, cohort, and scanner difference. A multi-type parallel feature embedding framework was trained to classify three FTD subtypes with a weighted cross-entropy loss function used to account for unbalanced sample sizes. Feature visualization was achieved through post-hoc analysis using an integrated gradient approach. The proposed differential diagnosis framework achieved a mean balanced accuracy of 0.80 for bvFTD, 0.82 for nfvPPA, 0.89 for svPPA, and an overall balanced accuracy of 0.84. Feature importance maps showed more localized differential patterns among different FTD subtypes compared to groupwise statistical mapping. In this study, we demonstrated the efficiency and effectiveness of using explainable deep-learning-based parallel feature embedding and visualization framework on MRI-derived multi-type structural patterns to differentiate three clinically defined subphenotypes of FTD: bvFTD, nfvPPA, and svPPA, which could help with the identification of at-risk populations for early and precise diagnosis for intervention planning.

Maize yield prediction with trait-missing data via bipartite graph neural network.

The timely and accurate prediction of maize (Zea mays L.) yields prior to harvest is critical for food security and agricultural policy development. Currently, many researchers are using machine learning and deep learning to predict maize yields in specific regions with high accuracy. However, existing methods typically have two limitations. One is that they ignore the extensive correlation in maize planting data, such as the association of maize yields between adjacent planting locations and the combined effect of meteorological features and maize traits on maize yields. The other issue is that the performance of existing models may suffer significantly when some data in maize planting records is missing, or the samples are unbalanced. Therefore, this paper proposes an end-to-end bipartite graph neural network-based model for trait data imputation and yield prediction. The maize planting data is initially converted to a bipartite graph data structure. Then, a yield prediction model based on a bipartite graph neural network is developed to impute missing trait data and predict maize yield. This model can mine correlations between different samples of data, correlations between different meteorological features and traits, and correlations between different traits. Finally, to address the issue of unbalanced sample size at each planting location, we propose a loss function based on the gradient balancing mechanism that effectively reduces the impact of data imbalance on the prediction model. When compared to other data imputation and prediction models, our method achieves the best yield prediction result even when missing data is not pre-processed.

Analysis of two binomial proportions in noninferiority confirmatory trials.

In this article, we propose considering an approximate exact score (AES) test for noninferiority comparisons and we derive its test-based confidence interval for the difference between two independent binomial proportions. This test was published in the literature, but not its associated confidence interval. The p-value for this test is obtained by using exact binomial probabilities with the nuisance parameter being replaced by its restricted maximum likelihood estimate. Calculated type I errors revealed that the AES method has important advantages for noninferiority comparisons over popular asymptotic methods for adequately powered confirmatory clinical trials, at 80% or 90% statistical power. For unbalanced sample sizes of the compared groups, type I errors for the asymptotic score method were shown to be higher than the nominal level in a systematic pattern over a range of true proportions, but the AES method did not suffer from such a problem. On average, the true type I error of the AES method was closer to the nominal level than all considered methods in the empirical comparisons. In rare cases, type I errors of the AES test exceeded the nominal level, but only by a small amount. Presented examples showed that the AES method can be more attractive in practice than practical exact methods. In addition, p-value and confidence interval of the AES method can be obtained in <30 s of computer time for most confirmatory trials. Theoretical arguments, combined with empirical evidence and fast computation time should make the AES method attractive in statistical practice.

A study on the sample representativeness of the Wisconsin Alzheimer’s Disease Research Center Clinical Core participants versus the Wisconsin state population

AbstractBackgroundUnderstanding how representative a research sample is compared to the population can help inform future recruitment planning, data analysis strategies, and interpretation of the generalizability of the findings. We compared the demographics, rural residence, neighborhood socioeconomic disadvantage, and vascular risk factors between the Wisconsin Alzheimer’s Disease Research Center (ADRC) Clinical Core participants versus the Wisconsin state adult population.MethodThe ADRC participant sample (n = 678) included all active participants as of November 30, 2021. The state population statistics were estimated using multiple national and state data sources. Differences between the sample versus the population were tested using the one‐sample z‐test, the one‐sample Wilcoxon signed rank test, the exact binomial test, and the exact Monte Carlo multinomial test. Except for age, comparisons were performed for two age groups: 45‐64 years and ≥ 65 years, separately.ResultThe ADRC participants are older than the state population. For each age group, compared to the state population, the ADRC participants include more women and greater proportions of Black and American Indians, have higher levels of educational attainment, have smaller percentages living in rural areas and more disadvantaged neighborhoods, and show lower vascular risks. The sample versus population differences are greater for the ≥ 65 years group (the group at higher risk for AD) than the 45‐64 years group for most metrics, however, are smaller for sex and neighborhood disadvantage.ConclusionThe ADRC targets recruiting middle‐aged and older adults with dementia, mild cognitive impairment, or increased risk for AD. Thus, the participants are not a suitable sample for epidemiology studies on AD prevalence. Oversampling underrepresented racial groups would alleviate the lack of statistical power associated with small or unbalanced sample sizes and increase the precision in effect size estimates for studies on racial disparities. Future recruitment should endeavor to include participants from a broader range of educational backgrounds and more residents in rural areas and disadvantaged neighborhoods. These subpopulations have inequitable access to quality health care and are at greater risk. Properly representing them in research would allow conclusions to be generalizable to these subpopulations and support policy making to meet their needs.

Differentiation of frontotemporal dementia subtypes using neuroimaging‐based multi‐type parallel feature embedding

AbstractBackgroundFrontotemporal dementia (FTD) represents a collection of neurocognitive syndromes with frontotemporal lobar degeneration (FTLD) neuropathology, and is associated with significant clinical, pathological, and genetic heterogeneity. We trained a deep neural network (DNN) classifier to differentiate behavioral‐variant FTD (bvFTD), semantic variant primary progressive aphasia (svPPA), and non‐fluent variant (nfvPPA) patients using MRI scans drawn from two multi‐site neuroimaging consortiums.MethodBvFTD (N = 173), nfvPPA (N = 63), and svPPA (N = 41) patients with T1‐MRI were extracted from the FTLD Neuroimaging Initiative (FTLDNI) and the ARTFL‐LEFFTDS Longitudinal Frontotemporal Lobar Degeneration (ALLFTD) databases. MRI data were preprocessed with FreeSurfer, with cortical thickness, cortical volume, and subcortical volumes extracted. Cortical measures were parcellated into 360 patches (i.e., ROIs) according to the HCP‐MMP1 atlas. Both the patch‐based cortical thickness and volume features were harmonized to control confounding effects of sex, age, total intracranial volume (TIV), cohort, and scanner. Multi‐type features were parallelly fed into a multi‐layer‐perceptron (MLP)‐based classifier. Weighted cross‐entropy loss function was used to account for unbalanced sample sizes across FTD subtypes. 10‐fold nested cross‐validation was used to evaluate the robustness of the classification model, with data split into 80/10/10 of training/validation/test data sets.ResultVisual evaluation z‐scores of cortical features among FTLDNI and ALLFTD groups revealed site‐specific differences significantly reduced by feature harmonization, especially for volume (Figure 1). The balanced accuracy of the ensembled DNN classifier of each FTD subtype on test sets across all ten folds reached 0.76 ± 0.09 for bvFTD, 0.79 ± 0.07 for nfvPPA, and 0.88 ± 0.07 for svPPA (Figure 2). Figure 3 shows confusion matrices for classification performance in the test set across ten cross‐validation folds.ConclusionWe developed a deep‐learning‐based framework to classify three FTD subtypes: bvFTD, nfvPPA, and svPPA. The combination of feature harmonization and parallel multi‐type feature embedding framework showed promising differentiation power. This work can be used to recognize at‐risk populations for early and precise diagnosis, to aid intervention planning. Future studies will compare classification based on different input features and use visualization methods to identify the most discriminative regions and explore their clinical relevance.

Detecting Multidimensional DIF in Polytomous Items with IRT Methods and Estimation Approaches

AbstractThe purpose of this study was to investigate multidimensional DIF with a simple and nonsimple structure in the context of multidimensional Graded Response Model (MGRM). This study examined and compared the performance of the IRT‐LR and Wald test using MML‐EM and MHRM estimation approaches with different test factors and test structures in simulation studies and applying real data sets. When the test structure included two dimensions, the IRT‐LR (MML‐EM) generally performed better than the Wald test and provided higher power rates. If the test included three dimensions, the methods provided similar performance in DIF detection. In contrast to these results, when the number of dimensions in the test was four, MML‐EM estimation completely lost precision in estimating the nonuniform DIF, even with large sample sizes. The Wald with MHRM estimation approaches outperformed the Wald test (MML‐EM) and IRT‐LR (MML‐EM). The Wald test had higher power rate and acceptable type I error rates for nonuniform DIF with the MHRM estimation approach.The small and/or unbalanced sample sizes, small DIF magnitudes, unequal ability distributions between groups, number of dimensions, estimation methods and test structure were evaluated as important test factors for detecting multidimensional DIF.

Diffusion Models-based Data Augmentation for the Cell Cycle Phase Classification

For biological research, sample size imbalance is common due to the nature of the research subjects. For example, in the study of the cell cycle phase, the sample size of dividing cells is also much smaller due to the extremely short duration of the mitotic phase compared to the interphase. Data augmentation using image generative models is an excellent way to address insufficient sample size and imbalanced distribution. In addition to the GAN-like models that have been extensively applied, the diffusion model, as an emerging model, has shown extraordinary performance in the field of image generation. This experiment uses the diffusion model as a means of image data enhancement. The experimental results expose that the performance of the classifier with data augmentation is significantly improved compared with the original dataset, and the positive predictive value is increased from about 0.7 to more than 0.9. The results reveal that the diffusion model has a good application prospect in the area of data enhancement and can effectively solve the problem of insufficient data or unbalanced sample size.

Application of Deep Learning in Image Recognition of Citrus Pests

The occurrence of pests at high frequencies has been identified as a major cause of reduced citrus yields, and early detection and prevention are of great significance to pest control. At present, studies related to citrus pest identification using deep learning suffer from unbalanced sample sizes between data set classes, which may cause slow convergence of network models and low identification accuracy. To address the above problems, this study built a dataset including 5182 pest images in 14 categories. Firstly, we expanded the dataset to 21,000 images by using the Attentive Recurrent Generative Adversarial Network (AR-GAN) data augmentation technique, then we built Visual Geometry Group Network (VGG), Residual Neural Network (ResNet) and MobileNet citrus pest recognition models by using transfer learning, and finally, we introduced an appropriate attention mechanism according to the model characteristics to enhance the ability of the three models to operate effectively in complex, real environments with greater emphasis placed on incorporating the deep features of the pests themselves. The results showed that the average recognition accuracy of the three models reached 93.65%, the average precision reached 93.82%, the average recall reached 93.65%, and the average F1-score reached 93.62%. The integrated application of data augmentation, transfer learning and attention mechanisms in the research can significantly enhance the model’s ability to classify citrus pests while saving training cost and time, which can be a reference for researchers in the industry or other fields.

Microbial and epidemiological factors in early detection of esophageal squamous cell carcinoma and precancerous lesions.

Microbial and epidemiological factors in early detection of esophageal squamous cell carcinoma and precancerous lesions.

Human liver microbiota modeling strategy at the early onset of fibrosis

BackgroundGut microbiota is involved in the development of liver diseases such as fibrosis. We and others identified that selected sets of gut bacterial DNA and bacteria translocate to tissues, notably the liver, to establish a non-infectious tissue microbiota composed of microbial DNA and a low frequency live bacteria. However, the precise set of bacterial DNA, and thereby the corresponding taxa associated with the early stages of fibrosis need to be identified. Furthermore, to overcome the impact of different group size and patient origins we adapted innovative statistical approaches. Liver samples with low liver fibrosis scores (F0, F1, F2), to study the early stages of the disease, were collected from Romania(n = 36), Austria(n = 10), Italy(n = 19), and Spain(n = 17). The 16S rRNA gene was sequenced. We considered the frequency, sparsity, unbalanced sample size between cohorts to identify taxonomic profiles and statistical differences.ResultsMultivariate analyses, including adapted spectral clustering with L1-penalty fair-discriminant strategies, and predicted metagenomics were used to identify that 50% of liver taxa associated with the early stage fibrosis were Enterobacteriaceae, Pseudomonadaceae, Xanthobacteriaceae and Burkholderiaceae. The Flavobacteriaceae and Xanthobacteriaceae discriminated between F0 and F1. Predicted metagenomics analysis identified that the preQ0 biosynthesis and the potential pathways involving glucoryranose and glycogen degradation were negatively associated with liver fibrosis F1-F2 vs F0.ConclusionsWithout demonstrating causality, our results suggest first a role of bacterial translocation to the liver in the progression of fibrosis, notably at the earliest stages. Second, our statistical approach can identify microbial signatures and overcome issues regarding sample size differences, the impact of environment, and sets of analyses.Trial registrationTirguMECCH ROLIVER Prospective Cohort for the Identification of Liver Microbiota, registration 4065/2014. Registered 01 01 2014.

Robust tests for multivariate repeated measures with small samples

Multivariate repeated measures data naturally arise in clinical trials and other fields such as biomedical science, public health, agriculture, social science and so on. For data of this type, the classical approach is to conduct multivariate analysis of variance (MANOVA) based on Wilks' Lambda and other multivariate statistics, which require the assumptions of multivariate normality and homogeneity of within-cell covariance matrices. However, data being analyzed nowadays show marked departure from multivariate normality and homogeneity. This paper proposes a finite-sample test by modifying the sums of squares matrices to make them insensitive to the heterogeneity in MANOVA. The proposed test is invariant to affine transformation and robust against nonnormality. The proposed method can be used in various experimental designs, for example, factorial design and crossover design. Under various simulation settings, the proposed method outperforms the classical Doubly Multivariate Model and Multivariate Mixed Model proposed elsewhere, especially for unbalanced sample sizes with heteroscedasticity. The applications of the proposed method are illustrated with ophthalmology data in factorial and crossover designs. The proposed method successfully identified and validated a significant main effect and demonstrated that univariate analysis could be oversensitive to small but clinically unimportant interactions.

Should studies with no events in both arms be excluded in evidence synthesis?

ObjectivesIn safety assessment, studies with no events are a frequent occurrence when conducting meta-analyses. The current approach in meta-analysis is to exclude double-zero studies from the synthesis. In this study, we compared the performance of excluding and including double-zero studies. MethodsA simulation with 5000 iterations was conducted based on the real-world dataset from Cochrane reviews. The true distribution of the rare events rather than normal distribution for the effects were used in the data generating mechanism to simulate aggregate meta-analysis data. We used Doi's inverse variance heterogeneity (IVhet) model for the meta-analyses with continuity correction (of 0.5) to include double-zero studies and used the odds ratio effect size. The performance of including versus excluding double-zero studies were then compared. ResultsGenerally, there was much larger mean squared error when double zero studies were excluded than when double-zero studies were included. The coverage when studies were excluded rapidly deteriorates as heterogeneity increased, while remained at or above the nominal level when double-zero studies were included. When there were very few double-zero studies, the performances was almost the same when including or excluding these studies. Subgroup analysis showed that, even for meta-analyses with unbalanced sample size across the two arms, including double-zero studies improved performance compared to when they were excluded. ConclusionsIncluding double-zero studies in meta-analysis improved performance substantively when compared to excluding them, especially when the proportion of double-zero studies was large. Continuity correction with use of the IVhet model is therefore a good solution to deal with double-zero studies and should be considered in future meta-analyses.

An Exact and Near-Exact Distribution Approach to the Behrens–Fisher Problem

The Behrens–Fisher problem occurs when testing the equality of means of two normal distributions without the assumption that the two variances are equal. This paper presents approaches based on the exact and near-exact distributions for the test statistic of the Behrens–Fisher problem, depending on different combinations of even or odd sample sizes. We present the exact distribution when both sample sizes are odd and the near-exact distribution when one or both sample sizes are even. The near-exact distributions are based on a finite mixture of generalized integer gamma (GIG) distributions, used as an approximation to the exact distribution, which consists of an infinite series. The proposed tests, based on the exact and the near-exact distributions, are compared with Welch’s t-test through Monte Carlo simulations, in particular for small and unbalanced sample sizes. The results show that the proposed approaches are competent solutions to the Behrens–Fisher problem, exhibiting precise sizes and better powers than Welch’s approach for those cases. Numerical studies show that the Welch’s t-test tends to be a bit more conservative than the test statistics based on the exact or near-exact distribution, in particular when sample sizes are small and unbalanced, situations in which the proposed exact or near-exact distributions obtain higher powers than Welch’s t-test.

Grouped-sampling technique to deal with unbalance in Raman spectral data modeling

Due to limitations in disease prevalence and hospital specificity, spectral data are often collected with unbalanced sample size. To solve this problem, a new sampling method – grouped-sampling was proposed in this research, which is shown to be effective for unbalanced data. It avoids over-fitting of over-sampling and overcomes under-sampling utilization of under-sampling. In this study, we applied grouped-sampling to two unbalanced datasets where the sample proportions are 199:40 and 75:225. And then verified from two classic models: PCA-SVM (Principal Component Analysis-Support Vector Machine) and the deep learning algorithm GoogLeNet. The accuracy of these two datasets were 85.11% and 96.15% in PCA-SVM and 85.10% and 84.61% on GoogLeNet. Also, the F1-score were evaluated to measure the classification balance of sampling method, and result shows that F1-score of grouped-sampling is always the highest compared to over-sampling and under-sampling. In summary, compared to traditional sampling methods, grouped-sampling performs better on prediction for classes with smaller sample size, which means grouped-sampling can improve the balance of classification results and the potential of practical application. Therefore, we develop a group sampling method that distinguishes between under- and over-sampling, which greatly improves the accuracy and balance of predictions for unbalanced samples.

Deep meta-learning and variational autoencoder for coupling fault diagnosis of rolling bearing under variable working conditions

Considering the characteristics of rolling bearing such as variable working conditions, unbalanced fault sample size, and multiple coupling fault types, it is a great challenge to achieve general accurate fault diagnosis model. In this paper, deep meta-learning and variational autoencoder (DML-VAE) is applied for coupling fault diagnosis of rolling bearing under variable working conditions. The collected vibration signals of rolling bearing are divided into long time series samples, including normal samples, single fault samples, and coupling fault samples. Then, variational autoencoder (VAE) is utilized for data augmentation of time series samples, and the generated samples are brought into one-dimensional deep convolutional neural network (1-DCNN) for further classification of multiple coupling faults. Subsequently, the trained 1-DCNN is regarded as embedding model. Training samples and other working condition samples are defined as support set and query set. Based on metric-based meta-learning method, sample pairs composed of support set and query set are constructed and brought into the embedding model to get the category with shortest metric distance as the classification result. In addition, the embedding model can be optimized by minimizing the contrastive loss among these sample pairs. The case study shows that the DML-VAE can achieve accurate classification results under the coupling of two faults and three faults, and maintain high diagnostic accuracy under variable working conditions. Compared with other models, the proposed model can also get the most accurate fault diagnosis results for all categories under unbalanced samples.

Abstract P2-01-03: Elucidating the biology of circulating tumor DNA (ctDNA) shedding across receptor subtypes in high-risk early-stage breast cancer

Abstract Background: Identifying mechanisms that govern the shedding of ctDNA in blood could inform the use of liquid biopsy in individual patients. Previous studies in the I-SPY2 neoadjuvant trial involving high-risk breast cancer showed that the detection of ctDNA before treatment was associated with aggressive clinical characteristics and residual ctDNA after treatment was associated with poor outcomes. Moreover, ctDNA positivity rates significantly varied across breast cancer subtypes suggesting that ctDNA shedding may in part be driven by subtype-specific etiology. We performed genome-wide transcriptomic analysis to identify genes and biological processes associated with increased ctDNA shedding within and across receptor subtypes. Methods: Our study involved 227 patients in I-SPY2 with tumor gene expression and ctDNA data at pretreatment. All patients were at high risk for recurrence (MammaPrint high). Each subtype: HR+HER2- (n=109), HER2+ (n=19), and triple negative breast cancer (TNBC, n=99) was evaluated independently. We performed differential expression (DE) analysis on the global transcriptome (m=19,134 genes) and curated gene signature (cGS, m=31 signatures developed in I-SPY2) data between ctDNA+ and ctDNA- patients at baseline. Gene-set enrichment analysis (GSEA) was also performed across hallmark (H, m=50), canonical pathway (CP, m=5,501), gene ontology (GO, m=9,996) and immunologic (IM, m=4,872) gene sets. Features were associated with ctDNA shedding if Benjamini-Hochberg adjusted p &amp;lt; 0.05. For subtypes with smaller sample size and unbalanced groups, we also report features with nominally significant p &amp;lt; 0.05. Results: ctDNA positivity rate was significantly higher in TNBC (91%) than in HR+HER2- and HER2+ (65% and 74% respectively, Fisher p &amp;lt;0.001). The HR+HER2- subtype had the most significant hits for DE analysis between ctDNA+ and ctDNA- patients, with 0.2% of genes and 3.2% of cGS. No genes or cGS were differentially expressed in TNBC and HER2+, likely due to imbalance or small size of these groups. For GSEA, we observed the most significant number of enrichments in HR+HER2- subtype, with 58%, 21.8%, 4.4% and 40.3% of H, CP, GO, and IM gene sets enriched, respectively. In the HER2+ subtype, 40% H, 15.7% CP and 36.4% IM gene sets were significantly enriched, while no gene sets were enriched in TNBC. To identify common mechanistic themes across subtypes, we also considered nominally significant features in DE and GSEA. Processes associated with infection and innate immune responses were enriched in ctDNA+ patients, while adaptive immune response and antigen presentation—e.g., T-cell, TCR and MHC II protein complex, and downregulation of MYC targets were enriched in ctDNA- patients. HR+HER2- and HER2+ subtypes shared the most common modulated features with 134 genes and 2,165 gene sets, including up-regulation of cell cycle and proliferation in ctDNA+ patients, as well as up- or down-regulation of specific immunologic and metabolic processes. In contrast, TNBC gene set enrichment was associated with more distinct biologic processes, sharing common enrichment of 113 and 27 gene sets with HR+HER2- and HER2+ subtype, respectively. Conclusions: Findings from our exploratory analysis suggest a key role of immune response pathways in the control of ctDNA release. Additionally, tumor cell proliferation was associated with increased shedding in HR+HER2- and HER2+ subtypes, while down regulation of MYC targets was associated with ctDNA- patients across all subtypes. These suggest an important role of cell cycle in ctDNA shedding. Overall, our analysis revealed common and unique mechanisms potentially associated with ctDNA shedding across and within subtypes. However, due to the unbalanced groups and limited sample sizes, validation in a larger cohort is warranted. Citation Format: Rosalyn W. Sayaman, Denise M. Wolf, Christina Yau, Lamorna Brown Swigart, Gillian L. Hirst, Laura Sit, Nicholas O'Grady, Amy L. Delson, I-SPY2 Investigators, Laura Esserman, Laura J. van 't Veer, Mark Jesus M. Magbanua. Elucidating the biology of circulating tumor DNA (ctDNA) shedding across receptor subtypes in high-risk early-stage breast cancer [abstract]. In: Proceedings of the 2021 San Antonio Breast Cancer Symposium; 2021 Dec 7-10; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2022;82(4 Suppl):Abstract nr P2-01-03.

A Study on the Teaching Design of a Hybrid Civics Course Based on the Improved Attention Mechanism

As an important vehicle for moral education, the moral indicators of civics and political science textbooks are naturally some of the most important criteria for revising textbooks. However, the textbook text dataset has too much textual information, ambiguous features, unbalanced sample distributions, etc. To address these problems, this paper combines a novel data enhancement method to obtain classification results based on word vectors. Additionally, for the problem of unbalanced sample sizes, this paper proposes a network model based on the attention mechanism, which combines the ideas of SMOTE and EDA, and uses a self-built stop word list and synonym word forest to conduct synonym queries, achieve a few categories of oversampling, and randomly disrupt the sentence order and intra-sentence word order to build a balanced dataset. The experimental results also show that the data augmentation method used in this paper’s model can effectively improve the performance of the model, resulting in a higher boost in the F1-measure of the model. The model incorporating the attention mechanism has better model generalization compared to the one without the attention mechanism, as well as a significant advantage compared to the reference model in other settings. The experimental results show that, compared with the original text classifier, the scheme of this paper effectively improves the evaluation effect and the reliability design for teaching a civics course.