GraphMHC: Neoantigen prediction model applying the graph neural network to molecular structure.

Neoantigens are tumor-derived peptides and are biomarkers that can predict prognosis related to immune checkpoint inhibition by estimating their binding to major histocompatibility complex (MHC) proteins. Although deep neural networks have been primarily used for these prediction models, it is difficult to interpret the models reported thus far as accurately representing the interactions between biomolecules. In this study, we propose the GraphMHC model, which utilizes a graph neural network model applied to molecular structure to simulate the binding between MHC proteins and peptide sequences. Amino acid sequences sourced from the immune epitope database (IEDB) undergo conversion into molecular structures. Subsequently, atomic intrinsic informations and inter-atomic connections are extracted and structured as a graph representation. Stacked graph attention and convolution layers comprise the GraphMHC network which classifies bindings. The prediction results from the test set using the GraphMHC model showed a high performance with an area under the receiver operating characteristic curve of 92.2% (91.9-92.5%), surpassing a baseline model. Moreover, by applying the GraphMHC model to melanoma patient data from The Cancer Genome Atlas project, we found a borderline difference (0.061) in overall survival and a significant difference in stromal score between the high and low neoantigen load groups. This distinction was not present in the baseline model. This study presents the first feature-intrinsic method based on biochemical molecular structure for modeling the binding between MHC protein sequences and neoantigen candidate peptide sequences. This model can provide highly accurate responsibility information that can predict the prognosis of immune checkpoint inhibitors to cancer patients who want to apply it.

Comparison of Newly Proposed LDL-Cholesterol Estimation Equations.

BACKGROUND: Low-density lipoprotein cholesterol is an important marker highly associated with cardiovascular disease. Since the direct measurement of it is inefficient in terms of cost and time, it is common to estimate through the Friedewald equation developed about 50 years ago. However, various limitations exist since the Friedewald equation was not designed for Koreans. This study proposes a new low-density lipoprotein cholesterol estimation equation for South Koreans using nationally approved statistical data. METHODS: This study used data from the Korean National Health and Nutrition Examination Survey from 2009 to 2019. The 18,837 subjects were used to develop the equation for estimating low-density lipoprotein cholesterol. The subjects included individuals with low-density lipoprotein cholesterol levels directly measured among those with high-density lipoprotein cholesterol, triglycerides, and total cholesterol measured. We compared twelve equations developed in the previous studies and the newly proposed equation (model 1) developed in this study with the actual low-density lipoprotein cholesterol value in various ways. RESULTS: The low-density lipoprotein cholesterol value estimated using the estimation formula and the actual low-density lipoprotein cholesterol value were compared using the root mean squared error. When the triglyceride level was less than 400 mg/dL, the root mean squared of the model 1 was 7.96, the lowest compared to other equations, and the model 2 was 7.82. The degree of misclassification was checked according to the NECP ATP III 6 categories. As a result, the misclassification rate of the model 1 was the lowest at 18.9%, and Weighted Kappa was the highest at 0.919 (0.003), which means it significantly reduced the underestimation rate shown in other existing estimation equations. Root mean square error was also compared according to the change in triglycerides level. As the triglycerides level increased, the root mean square error showed an increasing trend in all equations, but it was confirmed that the model 1 was the lowest compared to other equations. CONCLUSION: The newly proposed low-density lipoprotein cholesterol estimation equation showed significantly improved performance compared to the 12 existing estimation equations. The use of representative samples and external verification is required for more sophisticated estimates in the future.

Evaluation of an air quality warning system for vulnerable and susceptible individuals in Korea: an interrupted time series analysis.

OBJECTIVES: This study was conducted to elucidate the effects of an air quality warning system (AQWS) implemented in January 2015 in Korea by analyzing changes in the incidence and exacerbation rates of environmental diseases. METHODS: Data from patients with environmental diseases were extracted from the National Health Insurance Service-National Sample Cohort database from 2010 to 2019, and data on environmental risk factors were acquired from the AirKorea database. Patient and meteorological data were linked based on residential area. An interrupted time series analysis with Poisson segmented regression was used to compare the rates before and after AQWS introduction. Adjustment variables included seasonality, air pollutants (carbon monoxide, nitrogen dioxide, sulfur dioxide, particulate matter less than 10 µm in diameter, and ozone), temperature, and humidity. RESULTS: After AQWS implementation, the incidence of asthma gradually decreased by 20.5%. Cardiovascular disease and stroke incidence also significantly decreased (by 34.3 and 43.0%, respectively). However, no immediate or gradual decrease was identified in the exacerbation rate of any environmental disease after AQWS implementation. Sensitivity analyses were performed according to age, disability, and health insurance coverage type. Overall, the AQWS effectively mitigated the occurrence of most environmental diseases in Korea. However, the relationships between alarm system implementation and reduced incidence differed among diseases based on the characteristics of vulnerable and sensitive individuals. CONCLUSIONS: Our results suggest that by tailoring the AQWS to demographic and sociological characteristics and providing enhanced education about the warning system, interventions can become an efficient policy tool to decrease air pollution- related health risks.

Machine learning-based injury severity prediction of level 1 trauma center enrolled patients associated with car-to-car crashes in Korea.

Injury prediction models enables to improve trauma outcomes for motor vehicle occupants in accurate decision-making and early transport to appropriate trauma centers. This study aims to investigate the injury severity prediction (ISP) capability in machine-learning analytics based on five-different regional Level 1 trauma center enrolled patients in Korea. We study car crash-related injury data of 1417 patients enrolled in the Korea In-Depth Accident Study database from January 2011 to April 2021. Severe injury classification was defined using an Injury Severity Score of 15 or greater. A planar crash was considered by excluding rollovers to compromise an accurate prediction. Furthermore, dissimilarities of the collision partner component based on vehicle segmentation were assumed for crash incompatibility. To handle class-imbalanced clinical datasets, we used four data-sampling techniques (i.e., class-weighting, resampling, synthetic minority oversampling, and adaptive synthetic sampling). Machine-learning analytics based on logistic regression, extreme gradient boosting (XGBoost), and a multilayer perceptron model were used for the evaluations. Each model was executed using five-fold cross-validation to solve overfitting consistent with the hyperparameters tuned to improve model performance. The area under the receiver operating characteristic curve of 0.896. Additionally, the present ISP model showed an under-triage rate of 6.1%. The Delta-V, age, and Principal ~ were significant predictors. The results demonstrated that the data-balanced XGBoost model achieved a reliable performance on injury severity classification of emergency department patients. This finding considers ISP model selection, which affected prediction performance based on overall predictor variables.

Applications of deep learning methods in digital biomarker research using noninvasive sensing data.

Introduction: Noninvasive digital biomarkers are critical elements in digital healthcare in terms of not only the ease of measurement but also their use of raw data. In recent years, deep learning methods have been put to use to analyze these diverse heterogeneous data; these methods include representation learning for feature extraction and supervised learning for the prediction of these biomarkers. Methods: We introduce clinical cases of digital biomarkers and various deep-learning methods applied according to each data type. In addition, deep learning methods for the integrated analysis of multidimensional heterogeneous data are introduced, and the utility of these data as an integrated digital biomarker is presented. The current status of digital biomarker research is examined by surveying research cases applied to various types of data as well as modeling methods. Results: We present a future research direction for using data from heterogeneous sources together by introducing deep learning methods for dimensionality reduction and mode integration from multimodal digital biomarker studies covering related domains. The integration of multimodality has led to advances in research through the improvement of performance and complementarity between modes. Discussion: The integrative digital biomarker will be more useful for research on diseases that require data from multiple sources to be treated together. Since delicate signals from patients are not missed and the interaction effects between signals are also considered, it will be helpful for immediate detection and more accurate prediction of symptoms.

Prediction Model for Hypertension and Diabetes Mellitus Using Korean Public Health Examination Data (2002-2017).

Hypertension and diabetes mellitus are major chronic diseases that are important factors in the management of cardiovascular disease. In order to prevent the occurrence of chronic diseases, proper health management through periodic health check-ups is necessary. The purpose of this study is to determine the incidence of hypertension and diabetes mellitus according to the health check-up, and to develop a predictive model for hypertension and diabetes according to the health check-up. We used the National Health Insurance Corporation database of Korea and checked whether hypertension or diabetes occurred from that date according to the number of health check-ups over the past 10 years. Compared to those who underwent five health check-ups, those who participated in the first screening had hypertension (OR = 2.18, 95% CI = 2.14-2.22), diabetes mellitus (OR = 1.33, 95% CI = 1.30-1.35) and both diseases (OR = 2.46, 95% CI = 2.39-2.53); individuals who underwent 10 screenings had hypertension (OR = 0.86, 95% CI = 0.83-0.88), diabetes mellitus (OR = 0.83, 95% CI = 0.81-0.85) and both diseases (OR = 0.83, 95% CI = 0.79-0.87). Individuals who attended fewer than five screenings compared with individuals who attended five or more screenings had hypertension (OR = 1.61, 95% CI = 1.59-1.62; AUC = 0.66), diabetes mellitus (OR = 1.21, 95% CI = 1.20-1.22; AUC = 0.59) and both diseases (OR = 1.75, 95% CI = 1.72-1.78, AUC = 0.63). The machine learning-based prediction model using XGBoost showed higher performance in all datasets than the conventional logistic regression model in predicting hypertension (accuracy, 0.828 vs. 0.628; F1-score, 0.800 vs. 0.633; AUC, 828 vs. 0.630), diabetes mellitus (accuracy, 0.707 vs. 0.575; F1-score, 0.663 vs. 0.576; AUC, 0.710 vs. 0.575) and both diseases (accuracy, 0.950 vs. 0.612; F1-score, 0.950 vs. 0.614; AUC, 0.952 vs. 0.612). It was found that health check-up had a great influence on the occurrence of hypertension and diabetes, and screening frequency was more important than other factors in the variable importances.

Entropy-Based Graph Clustering of PPI Networks for Predicting Overlapping Functional Modules of Proteins.

Functional modules can be predicted using genome-wide protein-protein interactions (PPIs) from a systematic perspective. Various graph clustering algorithms have been applied to PPI networks for this task. In particular, the detection of overlapping clusters is necessary because a protein is involved in multiple functions under different conditions. graph entropy (GE) is a novel metric to assess the quality of clusters in a large, complex network. In this study, the unweighted and weighted GE algorithm is evaluated to prove the validity of predicting function modules. To measure clustering accuracy, the clustering results are compared to protein complexes and Gene Ontology (GO) annotations as references. We demonstrate that the GE algorithm is more accurate in overlapping clusters than the other competitive methods. Moreover, we confirm the biological feasibility of the proteins that occur most frequently in the set of identified clusters. Finally, novel proteins for the additional annotation of GO terms are revealed.

Is pelvic incidence a constant, as everyone knows? Changes of pelvic incidence in surgically corrected adult sagittal deformity.

PURPOSE: Previous investigations have recognized the critical role of pelvic parameters in the setting of a fixed sagittal deformity. Pelvic incidence (PI) is a constant, as everyone knows. However, PI might change reciprocally because of increased shear force on the sacroiliac joint, following surgical correction of fixed lumbar lordosis (LL). The disparity in PI after surgery according to the surgical method, and its impact on final follow-up, has not been reported. This study was undertaken to analyze the disparity of PI before and after surgery, and to evaluate its impact on final sagittal alignment in surgically corrected lordosis when there is immediate postoperative normal alignment following correction of adult sagittal deformity. METHODS: A prospective study of 29 subjects with adult spinal deformity (average age: 67.9 years) was conducted. At final evaluation after a minimum 2-year follow-up, normal sagittal alignment was achieved following consecutive sagittal correction. Surgical changes were measured by serial, pelvic standing, lateral, and whole spine radiographs, spinopelvic parameters measured included PI, sacral slope (SS), pelvic tilt (PT), LL, thoracic kyphosis (TK), and sagittal alignment. RESULTS: The mean LL was 0.2° before surgery; -59.3° after surgery with pedicle subtraction osteotomy (PSO) (n = 20), anterior lumbar interbody fusion (ALIF) (n = 20, 33 segments), and posterior lumbar interbody fusion (PLIF) (n = 21, 36 segments); and -57.5° at last follow-up. The sagittal vertical axis was +14.8 cm before surgery, -0.7 cm after surgery, and 2.2 cm at last follow-up. The mean PI was 49.4° before surgery, and increased to 55.2° after surgery, 57.5° at 1-year follow-up, and 58.8° at last follow-up (P = 0.02). The mean disparity in PI preoperatively and at last follow-up was 11.4° without sacropelvic fixation (n = 18), and 5.9° with sacropelvic fixation (n = 11) (P = 0.002). Analysis revealed the disparity of PI to be significantly greater in non-sacropelvic fixation, and correlated with the follow-up period (R = 0.442, P = 0.016), but not with age, bone mineral density (BMD), number of fused segments, correction methods, corrected LL, or sagittal alignment. CONCLUSIONS: PI increased in all patients with surgically corrected, adult sagittal deformity, following surgical correction of fixed LL. The disparity of PI after surgery was significantly higher in non-sacropelvic fixation, and showed a significant correlation with follow-up period without influence on sagittal alignment at last follow-up.