Meng Hao
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Research Interests
Meng Hao has a total of 61 co-authors. Here are the top co-authors by number of shared publications:
- Yi Li
Fudan University
9 shared publications - Hui Zhang
School of Global Health, Chinese Center for Tropical Diseases Research, Shanghai Jiao Tong University School of Medicine, Shanghai, China
6 shared publications Zixin Hu
Human Phenome Institute, Zhangjiang Fudan International Innovation Centre, Fudan University
6 shared publicationsLi Jin
Human Phenome Institute, Zhangjiang Fudan International Innovation Centre, Fudan University
5 shared publicationsXiangnan Li
Zhangjiang Fudan International Innovation Center, Human Phenome Institute, Fudan University, Shanghai, China
5 shared publicationsJingyi Wu
Fudan University
4 shared publications
Latest Publications
- Open Access
A Residual Approach to Estimate Biological Age from Gompertz Modeling
Biological age (BA) and its residual relative to chronological age are popularly used to quantify individual aging. Although these residuals independently predict age-related health outcomes, conventional BA measures often lack robustness in heterogeneous populations, and their residuals are not directly derivable in clinical practice. To address these limitations, we introduce the Gompertz law-based residual (GOLD-R) framework, a method designed to directly estimate BA residuals and optimized for cross-sectional data. We demonstrated the applicability and robustness of GOLD-R across multiple data types and populations. First, training on DNA methylation data from the EWAS Data Hub, the framework outperformed established epigenetic clocks in predicting mortality in a pan-cancer dataset. Then, applied to UK Biobank proteomics data, GOLD-R generated organismal and organ-specific aging measures that proved more robust than conventional age-prediction approaches in forecasting incident diseases and mortality. Finally, extending the analysis to clinical biomarkers using the NHANES and HRS, we found that GOLD-R residuals, derived from clinical biomarkers, surpassed those from both epigenetic and phenotypic clocks in performance. In summary, our findings establish GOLD-R as a robust algorithm for biological age estimation, providing a practical tool for both research and clinical applications.
Open AccessA Residual Approach to Estimate Biological Age from Gompertz Modeling
Biological age (BA) and its residual relative to chronological age are popularly used to quantify individual aging. Although these residuals independently predict age-related health outcomes, conventional BA measures often lack robustness in heterogeneous populations, and their residuals are not directly derivable in clinical practice. To address these limitations, we introduce the Gompertz law-based residual (GOLD-R) framework, a method designed to directly estimate BA residuals and optimized for cross-sectional data. We demonstrated the applicability and robustness of GOLD-R across multiple data types and populations. First, training on DNA methylation data from the EWAS Data Hub, the framework outperformed established epigenetic clocks in predicting mortality in a pan-cancer dataset. Then, applied to UK Biobank proteomics data, GOLD-R generated organismal and organ-specific aging measures that proved more robust than conventional age-prediction approaches in forecasting incident diseases and mortality. Finally, extending the analysis to clinical biomarkers using the NHANES and HRS, we found that GOLD-R residuals, derived from clinical biomarkers, surpassed those from both epigenetic and phenotypic clocks in performance. In summary, our findings establish GOLD-R as a robust algorithm for biological age estimation, providing a practical tool for both research and clinical applications.
- 20 September 2025
- Published by Springer Nature in GeroScience
- Vol. 48 (3) , 4195-4207
Social frailty and mortality risk in middle-aged and older adults: a prospective cohort study
Social frailty is common and associated with several adverse outcomes in older adults. However, its prevalence and effects on mortality in younger populations and the underlying cause of mortality are poorly understood. To examine the association of social frailty with all-cause and cause-specific mortality in adults across a wide age spectrum, we used data from the UK Biobank, a prospective cohort included 421,644 individuals aged 37–73 years enrolled from 2006 to 2010. Social frailty status was assessed based on Bunt’s concept (financial difficulty, live alone, less social activity and rarely contacts with friends/family). The prevalence of pre-social frailty and social frailty were 35.74% and 19.87%, respectively, indicating that more than half of the participants were at risk of cumulative depletion of essential social resources. Both pre-social frailty and social frailty were associated with higher risk of all-cause and cause-specific mortality (including malignant neoplasms, heart disease, cerebrovascular diseases, respiratory diseases, diabetes mellitus, and others caused) in adults across a wide age spectrum, independent of sociodemographic factors, lifestyles, chronic diseases, mental health, and physical frailty status. These findings indicate that social frailty, as a robust and multidimensional construct, effectively captures the risk of losing social resources. Thus, assessing and addressing social frailty can reduce mortality risk.
- 20 September 2025
- Published by Springer Nature in GeroScience
- Vol. 48 (3) , 4195-4207
Social frailty and mortality risk in middle-aged and older adults: a prospective cohort study
Social frailty is common and associated with several adverse outcomes in older adults. However, its prevalence and effects on mortality in younger populations and the underlying cause of mortality are poorly understood. To examine the association of social frailty with all-cause and cause-specific mortality in adults across a wide age spectrum, we used data from the UK Biobank, a prospective cohort included 421,644 individuals aged 37–73 years enrolled from 2006 to 2010. Social frailty status was assessed based on Bunt’s concept (financial difficulty, live alone, less social activity and rarely contacts with friends/family). The prevalence of pre-social frailty and social frailty were 35.74% and 19.87%, respectively, indicating that more than half of the participants were at risk of cumulative depletion of essential social resources. Both pre-social frailty and social frailty were associated with higher risk of all-cause and cause-specific mortality (including malignant neoplasms, heart disease, cerebrovascular diseases, respiratory diseases, diabetes mellitus, and others caused) in adults across a wide age spectrum, independent of sociodemographic factors, lifestyles, chronic diseases, mental health, and physical frailty status. These findings indicate that social frailty, as a robust and multidimensional construct, effectively captures the risk of losing social resources. Thus, assessing and addressing social frailty can reduce mortality risk.
- Open Access
Human aging reflects increases in entropy across organ networks
Aging involves diminished homeostatic control and changes in individual biomarker levels/states. However, it is unknown whether these alterations reflect a rise in entropy during the aging process, and whether entropy disrupts broad systemic interrelationships. The entropy of human aging has not been well characterized, but measures of systemic entropy could reveal aging dynamics that may not be apparent even by integration of state-based aging metrics. Here, we leverage the Distance of Covariance (DISCO), which quantifies entropy in large ensembles of biological information, to demonstrate that organs and systems exhibit interconnected increased entropy with age. We validate DISCO on multiple data substrates (clinical biomarkers, proteomics, metabolomics, and microbiomes) in five cohort datasets: UK BioBank, National Health and Nutrition Examination Survey, and three Chinese cohorts of older adults. DISCO consistently outperforms mortality prediction of existing metrics of dysregulation and is comparable to the best-in-class epigenetic clocks. It also strongly predicts frailty and incidence of age-related chronic conditions. Crucially, organ- and system-specific DISCO scores derived from circulating proteomics demonstrate broad predictive power with little to no specificity of a given organ predicting its own diseases and mortality. Network analysis of organ- and system-specific DISCO shows that more central, connected organ DISCOs predict health outcomes more strongly. For example, for each mortality cause, brain entropy is one of the strongest predictors. These findings challenge current notions of independent organ-specific aging signatures, suggesting instead that while pathology may be organ-specific, entropy spills readily across systems, and thus conversely that health during aging requires integrated homeostatic coordination across multiple systems.
Open AccessHuman aging reflects increases in entropy across organ networks
Aging involves diminished homeostatic control and changes in individual biomarker levels/states. However, it is unknown whether these alterations reflect a rise in entropy during the aging process, and whether entropy disrupts broad systemic interrelationships. The entropy of human aging has not been well characterized, but measures of systemic entropy could reveal aging dynamics that may not be apparent even by integration of state-based aging metrics. Here, we leverage the Distance of Covariance (DISCO), which quantifies entropy in large ensembles of biological information, to demonstrate that organs and systems exhibit interconnected increased entropy with age. We validate DISCO on multiple data substrates (clinical biomarkers, proteomics, metabolomics, and microbiomes) in five cohort datasets: UK BioBank, National Health and Nutrition Examination Survey, and three Chinese cohorts of older adults. DISCO consistently outperforms mortality prediction of existing metrics of dysregulation and is comparable to the best-in-class epigenetic clocks. It also strongly predicts frailty and incidence of age-related chronic conditions. Crucially, organ- and system-specific DISCO scores derived from circulating proteomics demonstrate broad predictive power with little to no specificity of a given organ predicting its own diseases and mortality. Network analysis of organ- and system-specific DISCO shows that more central, connected organ DISCOs predict health outcomes more strongly. For example, for each mortality cause, brain entropy is one of the strongest predictors. These findings challenge current notions of independent organ-specific aging signatures, suggesting instead that while pathology may be organ-specific, entropy spills readily across systems, and thus conversely that health during aging requires integrated homeostatic coordination across multiple systems.
- Open Access
Cardiovascular-Kidney-Metabolic Syndrome, Social Determinants of Health, and Dementia Risk: a prospective cohort study
Cardiovascular-kidney-metabolic (CKM) syndrome, characterized by pathophysiological interactions among metabolic risk factors, chronic kidney disease and the cardiovascular system, is a significant global health concern, particularly in populations with adverse social determinants of health (SDOH). However, the influence of CKM syndrome and the joint effects of SDOH on the risk of developing incident dementia has not been fully elucidated. Here, we examined these associations among 382569 individuals from the UK Biobank. We found that unfavorable SDOH and advanced CKM syndrome (stage 3-4) are independently associated with increased risks of incident all-cause dementia, Alzheimer’s disease and vascular dementia. Additionally, joint associations of CKM syndrome stages and SODH with incident dementia were observed. Individuals with ≥3 unfavorable SDOH and in CKM syndrome stage 4 exhibited the highest dementia risk, even after adjusting for APOE ε4 status. Our findings highlighted the importance of maintaining optimal CKM health and addressing unfavorable SODH in cognitive aging.
Open AccessCardiovascular-Kidney-Metabolic Syndrome, Social Determinants of Health, and Dementia Risk: a prospective cohort study
Cardiovascular-kidney-metabolic (CKM) syndrome, characterized by pathophysiological interactions among metabolic risk factors, chronic kidney disease and the cardiovascular system, is a significant global health concern, particularly in populations with adverse social determinants of health (SDOH). However, the influence of CKM syndrome and the joint effects of SDOH on the risk of developing incident dementia has not been fully elucidated. Here, we examined these associations among 382569 individuals from the UK Biobank. We found that unfavorable SDOH and advanced CKM syndrome (stage 3-4) are independently associated with increased risks of incident all-cause dementia, Alzheimer’s disease and vascular dementia. Additionally, joint associations of CKM syndrome stages and SODH with incident dementia were observed. Individuals with ≥3 unfavorable SDOH and in CKM syndrome stage 4 exhibited the highest dementia risk, even after adjusting for APOE ε4 status. Our findings highlighted the importance of maintaining optimal CKM health and addressing unfavorable SODH in cognitive aging.
- Top Cited PapersOpen Access
- 11 February 2025
- Published by Elsevier in JACC: Asia
- Vol. 5 (3) , 393-395
Prevalence of Cardiovascular-Kidney-Metabolic Syndrome Stages Among Middle-Aged and Older Adults in China
- 11 February 2025
- Published by Elsevier in JACC: Asia
- Vol. 5 (3) , 393-395
Top Cited PapersOpen AccessPrevalence of Cardiovascular-Kidney-Metabolic Syndrome Stages Among Middle-Aged and Older Adults in China
- Open Access
- 27 November 2024
- Published by Qeios Ltd in Qeios Preprints
Gompertz Law-Based Biological Age (GOLD BioAge): A Simple and Practical Measurement of Biological Aging to Capture Morbidity and Mortality Risks
Biological age reflects actual aging and overall health, but current aging clocks are often complex and difficult to interpret, limiting their clinical application. In this study, we introduced a Gompertz law-based biological age (GOLD BioAge) model that simplified aging assessment. We estimated GOLD BioAge using clinical biomarkers and found significant associations of the difference from chronological age (BioAgeDiff) with risks of morbidity and mortality in NHANES. Moreover, we developed GOLD ProtAge and MetAge using proteomics and metabolomics data, which outperformed the clinical-only model in predicting mortality and chronic disease risks in UK Biobank. Benchmark analysis illustrated that our models exceeded common aging clocks in predicting mortality across diverse age groups in both NHANES and UK Biobank. The results demonstrated that the GOLD BioAge algorithm effectively applied to both clinical and omics data, showing excellent performance in predicting age-related outcomes. Additionally, we created a simplified version called the Light BioAge, which used three biomarkers for aging assessment. The Light model reliably captured mortality risks in three validation cohorts (CHARLS, RuLAS, CLHLS). It significantly predicted the onset of frailty, stratified frail individuals, and collectively identified individuals at high risk of mortality. In summary, the algorithm of GOLD BioAge could provide a valuable framework for aging assessment in public health and clinical practice.
- 27 November 2024
- Published by Qeios Ltd in Qeios Preprints
Open AccessGompertz Law-Based Biological Age (GOLD BioAge): A Simple and Practical Measurement of Biological Aging to Capture Morbidity and Mortality Risks
Biological age reflects actual aging and overall health, but current aging clocks are often complex and difficult to interpret, limiting their clinical application. In this study, we introduced a Gompertz law-based biological age (GOLD BioAge) model that simplified aging assessment. We estimated GOLD BioAge using clinical biomarkers and found significant associations of the difference from chronological age (BioAgeDiff) with risks of morbidity and mortality in NHANES. Moreover, we developed GOLD ProtAge and MetAge using proteomics and metabolomics data, which outperformed the clinical-only model in predicting mortality and chronic disease risks in UK Biobank. Benchmark analysis illustrated that our models exceeded common aging clocks in predicting mortality across diverse age groups in both NHANES and UK Biobank. The results demonstrated that the GOLD BioAge algorithm effectively applied to both clinical and omics data, showing excellent performance in predicting age-related outcomes. Additionally, we created a simplified version called the Light BioAge, which used three biomarkers for aging assessment. The Light model reliably captured mortality risks in three validation cohorts (CHARLS, RuLAS, CLHLS). It significantly predicted the onset of frailty, stratified frail individuals, and collectively identified individuals at high risk of mortality. In summary, the algorithm of GOLD BioAge could provide a valuable framework for aging assessment in public health and clinical practice.
- Open Access
Gompertz law based biological age (GOLD BioAge): a simple and practical measurement of biological aging to capture morbidity and mortality risks
Biological age reflects actual aging and overall health, but current aging clocks are often complex and difficult to interpret, limiting their clinical application. In this study, we introduced a Gompertz law-based biological age (GOLD BioAge) model that simplified aging assessment. We estimated GOLD BioAge using clinical biomarkers and found significant associations of the difference from chronological age (BioAgeDiff) with risks of morbidity and mortality in NHANES. Moreover, we developed GOLD ProtAge and MetAge using proteomics and metabolomics data, which outperformed the clinical-only model in predicting mortality and chronic disease risks in UK Biobank. Benchmark analysis illustrated that our models exceeded common aging clocks in predicting mortality across diverse age groups in both NHANES and UK Biobank. The results demonstrated that the GOLD BioAge algorithm effectively applied to both clinical and omics data, showing excellent performance in predicting age-related outcomes. Additionally, we created a simplified version called the Light BioAge, which used three biomarkers for aging assessment. The Light model reliably captured mortality risks in three validation cohorts (CHARLS, RuLAS, CLHLS). It significantly predicted the onset of frailty, stratified frail individuals, and collectively identified individuals at high risk of mortality. In summary, the algorithm of GOLD BioAge could provide a valuable framework for aging assessment in public health and clinical practice. Highlights: The algorithm of Gompertz law based biological age (GOLD BioAge) was proposed to construct biological aging clocks with convenient and interpretable calculations, which had better performance in predicting mortality risks. Our approach was applicable to proteomics and metabolomics, yielding ProtAge and MetAge with great clinical prospect to improve accuracy of aging assessment and prevent age-related diseases. The Light BioAge, a simplified version, was developed using age and three biomarkers, and it independently predicted mortality in three cohorts. The Light BioAgeDiff significantly predicted the onset of frailty, stratified frail individuals, and collectively identified individuals at high risk of mortality.
Open AccessGompertz law based biological age (GOLD BioAge): a simple and practical measurement of biological aging to capture morbidity and mortality risks
Biological age reflects actual aging and overall health, but current aging clocks are often complex and difficult to interpret, limiting their clinical application. In this study, we introduced a Gompertz law-based biological age (GOLD BioAge) model that simplified aging assessment. We estimated GOLD BioAge using clinical biomarkers and found significant associations of the difference from chronological age (BioAgeDiff) with risks of morbidity and mortality in NHANES. Moreover, we developed GOLD ProtAge and MetAge using proteomics and metabolomics data, which outperformed the clinical-only model in predicting mortality and chronic disease risks in UK Biobank. Benchmark analysis illustrated that our models exceeded common aging clocks in predicting mortality across diverse age groups in both NHANES and UK Biobank. The results demonstrated that the GOLD BioAge algorithm effectively applied to both clinical and omics data, showing excellent performance in predicting age-related outcomes. Additionally, we created a simplified version called the Light BioAge, which used three biomarkers for aging assessment. The Light model reliably captured mortality risks in three validation cohorts (CHARLS, RuLAS, CLHLS). It significantly predicted the onset of frailty, stratified frail individuals, and collectively identified individuals at high risk of mortality. In summary, the algorithm of GOLD BioAge could provide a valuable framework for aging assessment in public health and clinical practice. Highlights: The algorithm of Gompertz law based biological age (GOLD BioAge) was proposed to construct biological aging clocks with convenient and interpretable calculations, which had better performance in predicting mortality risks. Our approach was applicable to proteomics and metabolomics, yielding ProtAge and MetAge with great clinical prospect to improve accuracy of aging assessment and prevent age-related diseases. The Light BioAge, a simplified version, was developed using age and three biomarkers, and it independently predicted mortality in three cohorts. The Light BioAgeDiff significantly predicted the onset of frailty, stratified frail individuals, and collectively identified individuals at high risk of mortality.
- Open Access
COVID-19 Epidemic Outside China: 34 Founders and Exponential Growth
Summary: Background: In December 2019, pneumonia infected with a novel coronavirus burst in Wuhan, China. Now the situation is almost controlled in China but is worse outside China. We aimed to build a mathematical model to capture the global trend of epidemics outside China. Methods: In this retrospective, outside-China diagnosis number reported from Jan 21 to Feb 28, 2020 was downloaded from WHO website. We develop a simple regression model on these numbers: where N t is the total diagnosed patient till the i th day, t=1 at Feb 1. Findings: Based on this model, we estimate that there have been about 34 unobserved founder patients at the beginning of spread outside China. The global trend is approximately exponential, with the rate of 10 folds every 19 days. Research in context: Evidence before this study: In December 2019, pneumonia infected with a novel coronavirus burst in Wuhan, China. Now the situation is almost controlled in China but is worse outside China. Now there are 4,691 patients across 51 countries and territories outside China. We searched PubMed and the China National Knowledge Infrastructure database for articles published up to Feb 28, 2020, using the keywords “COVID”, “novel coronavirus”, “2019-nCoV” or “2019 novel coronavirus”. No published work about the global trend of epidemics outside China could be identified. Added value of this study: We built a simple “log-plus” linear model to capture the global trend of epidemics outside China. We estimate that there have been about 34 unobserved founder patients at the beginning of spread outside China. The global trend is approximately exponential, with the rate of 10 folds every 19 days. Implications of all the available evidence: With the limited number of data points and the complexity of the real situation, a straightforward model is expected to work better. Our model suggests that the COVID-19 disease follows an approximate exponential growth model stably at the very beginning. We predict that the number of confirmed patients outside China will increase ten folds in every 19 days without strong intervention by applying our model. Powerful actions on public health should be taken to combat this epidemic all over the world.
Open AccessCOVID-19 Epidemic Outside China: 34 Founders and Exponential Growth
Summary: Background: In December 2019, pneumonia infected with a novel coronavirus burst in Wuhan, China. Now the situation is almost controlled in China but is worse outside China. We aimed to build a mathematical model to capture the global trend of epidemics outside China. Methods: In this retrospective, outside-China diagnosis number reported from Jan 21 to Feb 28, 2020 was downloaded from WHO website. We develop a simple regression model on these numbers: where N t is the total diagnosed patient till the i th day, t=1 at Feb 1. Findings: Based on this model, we estimate that there have been about 34 unobserved founder patients at the beginning of spread outside China. The global trend is approximately exponential, with the rate of 10 folds every 19 days. Research in context: Evidence before this study: In December 2019, pneumonia infected with a novel coronavirus burst in Wuhan, China. Now the situation is almost controlled in China but is worse outside China. Now there are 4,691 patients across 51 countries and territories outside China. We searched PubMed and the China National Knowledge Infrastructure database for articles published up to Feb 28, 2020, using the keywords “COVID”, “novel coronavirus”, “2019-nCoV” or “2019 novel coronavirus”. No published work about the global trend of epidemics outside China could be identified. Added value of this study: We built a simple “log-plus” linear model to capture the global trend of epidemics outside China. We estimate that there have been about 34 unobserved founder patients at the beginning of spread outside China. The global trend is approximately exponential, with the rate of 10 folds every 19 days. Implications of all the available evidence: With the limited number of data points and the complexity of the real situation, a straightforward model is expected to work better. Our model suggests that the COVID-19 disease follows an approximate exponential growth model stably at the very beginning. We predict that the number of confirmed patients outside China will increase ten folds in every 19 days without strong intervention by applying our model. Powerful actions on public health should be taken to combat this epidemic all over the world.
- Open Access
A Note on COVID-19 Diagnosis Number Prediction Model in China
Importance: To predict the diagnosed COVID-19 patients and the trend of the epidemic in China. It may give the public some scientific information to ease the fear of the epidemic. Objective: In December 2019, pneumonia infected with the novel coronavirus burst in Wuhan, China. We aimed to use a mathematical model to predict number of diagnosed patients in future to ease anxiety on the emergent situation. Design: According to all diagnosis number from WHO website and combining with the transmission mode of infectious diseases, the mathematical model was fitted to predict future trend of outbreak. Setting: Our model was based on the epidemic situation in China, which could provide referential significance for disease prediction in other countries, and provide clues for prevention and intervention of relevant health authorities. Participants: In this retrospective, all diagnosis number from Jan 21 to Feb 10, 2020 reported from China was included and downloaded from WHO website. Main Outcome(s) and Measure(s): We develop a simple but accurate formula to predict the next day diagnosis number: ,where N i is the total diagnosed patient till the i th day, and α was estimated as 0.904 at Feb 10. Results: Based on this model, it is predicted that the rate of disease infection will decrease exponentially. The total number of infected people is limited; thus, the disease will have limited impact. However, new diagnosis will last to March. Conclusions and Relevance: Through the establishment of our model, we can better predict the trend of the epidemic in China.
Open AccessA Note on COVID-19 Diagnosis Number Prediction Model in China
Importance: To predict the diagnosed COVID-19 patients and the trend of the epidemic in China. It may give the public some scientific information to ease the fear of the epidemic. Objective: In December 2019, pneumonia infected with the novel coronavirus burst in Wuhan, China. We aimed to use a mathematical model to predict number of diagnosed patients in future to ease anxiety on the emergent situation. Design: According to all diagnosis number from WHO website and combining with the transmission mode of infectious diseases, the mathematical model was fitted to predict future trend of outbreak. Setting: Our model was based on the epidemic situation in China, which could provide referential significance for disease prediction in other countries, and provide clues for prevention and intervention of relevant health authorities. Participants: In this retrospective, all diagnosis number from Jan 21 to Feb 10, 2020 reported from China was included and downloaded from WHO website. Main Outcome(s) and Measure(s): We develop a simple but accurate formula to predict the next day diagnosis number: ,where N i is the total diagnosed patient till the i th day, and α was estimated as 0.904 at Feb 10. Results: Based on this model, it is predicted that the rate of disease infection will decrease exponentially. The total number of infected people is limited; thus, the disease will have limited impact. However, new diagnosis will last to March. Conclusions and Relevance: Through the establishment of our model, we can better predict the trend of the epidemic in China.
- Open Access
- 22 November 2018
- Published by Springer Nature in BMC Bioinformatics
- Vol. 19 (1) , 1-12
knnAUC: an open-source R package for detecting nonlinear dependence between one continuous variable and one binary variable
Testing the dependence of two variables is one of the fundamental tasks in statistics. In this work, we developed an open-source R package (knnAUC) for detecting nonlinear dependence between one continuous variable X and one binary dependent variables Y (0 or 1). We addressed this problem by using knnAUC (k-nearest neighbors AUC test, the R package is available at https://sourceforge.net/projects/knnauc/ ). In the knnAUC software framework, we first resampled a dataset to get the training and testing dataset according to the sample ratio (from 0 to 1), and then constructed a k-nearest neighbors algorithm classifier to get the yhat estimator (the probability of y = 1) of testy (the true label of testing dataset). Finally, we calculated the AUC (area under the curve of receiver operating characteristic) estimator and tested whether the AUC estimator is greater than 0.5. To evaluate the advantages of knnAUC compared to seven other popular methods, we performed extensive simulations to explore the relationships between eight different methods and compared the false positive rates and statistical power using both simulated and real datasets (Chronic hepatitis B datasets and kidney cancer RNA-seq datasets). We concluded that knnAUC is an efficient R package to test non-linear dependence between one continuous variable and one binary dependent variable especially in computational biology area.
- 22 November 2018
- Published by Springer Nature in BMC Bioinformatics
- Vol. 19 (1) , 1-12
Open AccessknnAUC: an open-source R package for detecting nonlinear dependence between one continuous variable and one binary variable
Testing the dependence of two variables is one of the fundamental tasks in statistics. In this work, we developed an open-source R package (knnAUC) for detecting nonlinear dependence between one continuous variable X and one binary dependent variables Y (0 or 1). We addressed this problem by using knnAUC (k-nearest neighbors AUC test, the R package is available at https://sourceforge.net/projects/knnauc/ ). In the knnAUC software framework, we first resampled a dataset to get the training and testing dataset according to the sample ratio (from 0 to 1), and then constructed a k-nearest neighbors algorithm classifier to get the yhat estimator (the probability of y = 1) of testy (the true label of testing dataset). Finally, we calculated the AUC (area under the curve of receiver operating characteristic) estimator and tested whether the AUC estimator is greater than 0.5. To evaluate the advantages of knnAUC compared to seven other popular methods, we performed extensive simulations to explore the relationships between eight different methods and compared the false positive rates and statistical power using both simulated and real datasets (Chronic hepatitis B datasets and kidney cancer RNA-seq datasets). We concluded that knnAUC is an efficient R package to test non-linear dependence between one continuous variable and one binary dependent variable especially in computational biology area.