Xi Luo
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Xi Luo has a total of 98 co-authors. Here are the top co-authors by number of shared publications:
- Chiang-Shan Ray Li
Department of Psychiatry Yale University School of Medicine New Haven Connecticut USA
10 shared publications Sheng Zhang
Department of Psychiatry, Yale University, New Haven, CT 06519, USA
5 shared publicationsChiang-Shan R. Li
4 shared publications- Jaime Shinsuke Ide
Department of Psychiatry, Yale University School of Medicine, New Haven, Connecticut
4 shared publications Sien Hu
4 shared publications- Olivia M. Farr
Division of Endocrinology, Diabetes and Metabolism, Beth Israel Deaconess Medical Center, Harvard Medical School, Boston, MA 02215, USA
3 shared publications
Latest Publications
- Open Access
Graph Neural Networks with Transformer Fusion of Brain Connectivity Dynamics and Tabular Data for Forecasting Future Tobacco Use
Integrating non-Euclidean brain imaging data with Euclidean tabular data, such as clinical and demographic information, poses a substantial challenge for medical imaging analysis, particularly in forecasting future outcomes. While machine learning and deep learning techniques have been applied successfully to cross-sectional classification and prediction tasks, effectively forecasting outcomes in longitudinal imaging studies remains challenging. To address this challenge, we introduce a time-aware graph neural network model with transformer fusion (GNN-TF). This model flexibly integrates both tabular data and dynamic brain connectivity data, leveraging the temporal order of these variables within a coherent framework. By incorporating non-Euclidean and Euclidean sources of information from a longitudinal resting-state fMRI dataset from the National Consortium on Alcohol and Neurodevelopment in Adolescence (NCANDA), the GNN-TF enables a comprehensive analysis that captures critical aspects of longitudinal imaging data. Comparative analyses against a variety of established machine learning and deep learning models demonstrate that GNN-TF outperforms these state-of-the-art methods, delivering superior predictive accuracy for predicting future tobacco usage. The end-to-end, time-aware transformer fusion structure of the proposed GNN-TF model successfully integrates multiple data modalities and leverages temporal dynamics, making it a valuable analytic tool for functional brain imaging studies focused on clinical outcome prediction.
Open AccessGraph Neural Networks with Transformer Fusion of Brain Connectivity Dynamics and Tabular Data for Forecasting Future Tobacco Use
Integrating non-Euclidean brain imaging data with Euclidean tabular data, such as clinical and demographic information, poses a substantial challenge for medical imaging analysis, particularly in forecasting future outcomes. While machine learning and deep learning techniques have been applied successfully to cross-sectional classification and prediction tasks, effectively forecasting outcomes in longitudinal imaging studies remains challenging. To address this challenge, we introduce a time-aware graph neural network model with transformer fusion (GNN-TF). This model flexibly integrates both tabular data and dynamic brain connectivity data, leveraging the temporal order of these variables within a coherent framework. By incorporating non-Euclidean and Euclidean sources of information from a longitudinal resting-state fMRI dataset from the National Consortium on Alcohol and Neurodevelopment in Adolescence (NCANDA), the GNN-TF enables a comprehensive analysis that captures critical aspects of longitudinal imaging data. Comparative analyses against a variety of established machine learning and deep learning models demonstrate that GNN-TF outperforms these state-of-the-art methods, delivering superior predictive accuracy for predicting future tobacco usage. The end-to-end, time-aware transformer fusion structure of the proposed GNN-TF model successfully integrates multiple data modalities and leverages temporal dynamics, making it a valuable analytic tool for functional brain imaging studies focused on clinical outcome prediction.
- Open Access
- 17 August 2023
- Published by Frontiers Media SA in Frontiers in Big Data
- Vol. 6, 846202
Comorbidity network analysis using graphical models for electronic health records
Importance: The comorbidity network represents multiple diseases and their relationships in a graph. Understanding comorbidity networks among critical care unit (CCU) patients can help doctors diagnose patients faster, minimize missed diagnoses, and potentially decrease morbidity and mortality.Objective: The main objective of this study was to identify the comorbidity network among CCU patients using a novel application of a machine learning method (graphical modeling method). The second objective was to compare the machine learning method with a traditional pairwise method in simulation.Method: This cross-sectional study used CCU patients' data from Medical Information Mart for the Intensive Care-3 (MIMIC-3) dataset, an electronic health record (EHR) of patients with CCU hospitalizations within Beth Israel Deaconess Hospital from 2001 to 2012. A machine learning method (graphical modeling method) was applied to identify the comorbidity network of 654 diagnosis categories among 46,511 patients.Results: Out of the 654 diagnosis categories, the graphical modeling method identified a comorbidity network of 2,806 associations in 510 diagnosis categories. Two medical professionals reviewed the comorbidity network and confirmed that the associations were consistent with current medical understanding. Moreover, the strongest association in our network was between “poisoning by psychotropic agents” and “accidental poisoning by tranquilizers” (logOR 8.16), and the most connected diagnosis was “disorders of fluid, electrolyte, and acid–base balance” (63 associated diagnosis categories). Our method outperformed traditional pairwise comorbidity network methods in simulation studies. Some strongest associations between diagnosis categories were also identified, for example, “diagnoses of mitral and aortic valve” and “other rheumatic heart disease” (logOR: 5.15). Furthermore, our method identified diagnosis categories that were connected with most other diagnosis categories, for example, “disorders of fluid, electrolyte, and acid–base balance” was associated with 63 other diagnosis categories. Additionally, using a data-driven approach, our method partitioned the diagnosis categories into 14 modularity classes.Conclusion and relevance: Our graphical modeling method inferred a logical comorbidity network whose associations were consistent with current medical understanding and outperformed traditional network methods in simulation. Our comorbidity network method can potentially assist CCU doctors in diagnosing patients faster and minimizing missed diagnoses.
- 17 August 2023
- Published by Frontiers Media SA in Frontiers in Big Data
- Vol. 6, 846202
Open AccessComorbidity network analysis using graphical models for electronic health records
Importance: The comorbidity network represents multiple diseases and their relationships in a graph. Understanding comorbidity networks among critical care unit (CCU) patients can help doctors diagnose patients faster, minimize missed diagnoses, and potentially decrease morbidity and mortality.Objective: The main objective of this study was to identify the comorbidity network among CCU patients using a novel application of a machine learning method (graphical modeling method). The second objective was to compare the machine learning method with a traditional pairwise method in simulation.Method: This cross-sectional study used CCU patients' data from Medical Information Mart for the Intensive Care-3 (MIMIC-3) dataset, an electronic health record (EHR) of patients with CCU hospitalizations within Beth Israel Deaconess Hospital from 2001 to 2012. A machine learning method (graphical modeling method) was applied to identify the comorbidity network of 654 diagnosis categories among 46,511 patients.Results: Out of the 654 diagnosis categories, the graphical modeling method identified a comorbidity network of 2,806 associations in 510 diagnosis categories. Two medical professionals reviewed the comorbidity network and confirmed that the associations were consistent with current medical understanding. Moreover, the strongest association in our network was between “poisoning by psychotropic agents” and “accidental poisoning by tranquilizers” (logOR 8.16), and the most connected diagnosis was “disorders of fluid, electrolyte, and acid–base balance” (63 associated diagnosis categories). Our method outperformed traditional pairwise comorbidity network methods in simulation studies. Some strongest associations between diagnosis categories were also identified, for example, “diagnoses of mitral and aortic valve” and “other rheumatic heart disease” (logOR: 5.15). Furthermore, our method identified diagnosis categories that were connected with most other diagnosis categories, for example, “disorders of fluid, electrolyte, and acid–base balance” was associated with 63 other diagnosis categories. Additionally, using a data-driven approach, our method partitioned the diagnosis categories into 14 modularity classes.Conclusion and relevance: Our graphical modeling method inferred a logical comorbidity network whose associations were consistent with current medical understanding and outperformed traditional network methods in simulation. Our comorbidity network method can potentially assist CCU doctors in diagnosing patients faster and minimizing missed diagnoses.
- Open Access
- 20 April 2023
- Published by Wiley in Journal of Applied Clinical Medical Physics
- Vol. 24 (7) , e13970
Parametric delineation uncertainties contouring (PDUC) modeling on CT scans of prostate cancer patients
Purpose: Variability in contouring contributes to large variations in radiation therapy planning and treatment outcomes. The development and testing of tools to automatically detect contouring errors require a source of contours that includes well‐understood and realistic errors. The purpose of this work was to develop a simulation algorithm that intentionally injects errors of varying magnitudes into clinically accepted contours and produces realistic contours with different levels of variability.Methods: We used a dataset of CT scans from 14 prostate cancer patients with clinician‐drawn contours of the regions of interest (ROI) of the prostate, bladder, and rectum. Using our newly developed Parametric Delineation Uncertainties Contouring (PDUC) model, we automatically generated alternative, realistic contours. The PDUC model consists of the contrast‐based DU generator and a 3D smoothing layer. The DU generator transforms contours (deformation, contraction, and/or expansion) as a function of image contrast. The generated contours undergo 3D smoothing to obtain a realistic look. After model building, the first batch of auto‐generated contours was reviewed. Editing feedback from the reviews was then used in a filtering model for the auto‐selection of clinically acceptable (minor‐editing) DU contours.Results: Overall, C values of 5 and 50 consistently produced high proportions of minor‐editing contours across all ROI compared to the other C values (0.936 0.111 and 0.552 0.228, respectively). The model performed best on the bladder, which had the highest proportion of minor‐editing contours (0.606) of the three ROI. In addition, the classification AUC for the filtering model across all three ROI is 0.724 0.109.Discussion: The proposed methodology and subsequent results are promising and could have a great impact on treatment planning by generating mathematically simulated alternative structures that are clinically relevant and realistic enough (i.e., similar to clinician‐drawn contours) to be used in quality control of radiation therapy.
- 20 April 2023
- Published by Wiley in Journal of Applied Clinical Medical Physics
- Vol. 24 (7) , e13970
Open AccessParametric delineation uncertainties contouring (PDUC) modeling on CT scans of prostate cancer patients
Purpose: Variability in contouring contributes to large variations in radiation therapy planning and treatment outcomes. The development and testing of tools to automatically detect contouring errors require a source of contours that includes well‐understood and realistic errors. The purpose of this work was to develop a simulation algorithm that intentionally injects errors of varying magnitudes into clinically accepted contours and produces realistic contours with different levels of variability.Methods: We used a dataset of CT scans from 14 prostate cancer patients with clinician‐drawn contours of the regions of interest (ROI) of the prostate, bladder, and rectum. Using our newly developed Parametric Delineation Uncertainties Contouring (PDUC) model, we automatically generated alternative, realistic contours. The PDUC model consists of the contrast‐based DU generator and a 3D smoothing layer. The DU generator transforms contours (deformation, contraction, and/or expansion) as a function of image contrast. The generated contours undergo 3D smoothing to obtain a realistic look. After model building, the first batch of auto‐generated contours was reviewed. Editing feedback from the reviews was then used in a filtering model for the auto‐selection of clinically acceptable (minor‐editing) DU contours.Results: Overall, C values of 5 and 50 consistently produced high proportions of minor‐editing contours across all ROI compared to the other C values (0.936 0.111 and 0.552 0.228, respectively). The model performed best on the bladder, which had the highest proportion of minor‐editing contours (0.606) of the three ROI. In addition, the classification AUC for the filtering model across all three ROI is 0.724 0.109.Discussion: The proposed methodology and subsequent results are promising and could have a great impact on treatment planning by generating mathematically simulated alternative structures that are clinically relevant and realistic enough (i.e., similar to clinician‐drawn contours) to be used in quality control of radiation therapy.
- 12 July 2022
- Published by Wiley in Addiction Biology
- Vol. 27 (5) , e13208
Alcohol and cannabis co‐use and longitudinal gray matter volumetric changes in early and late adolescence
Background: Previous studies have characterized the impact of substance use on cerebral structure and function in adolescents. Yet, the great majority of prior studies employed a small sample, presented cross‐sectional findings, and omitted potential sex differences.Methods: Using data based on 724 adolescents (370 females) curated from the NCANDA study, we investigated how gray matter volumes (GMVs) decline longitudinally as a result of alcohol and cannabis use. The impacts of alcohol and cannabis co‐use and how these vary across assigned sex at birth and age were examined. Brain imaging data comprised the GMVs of 34 regions of interest and the results were evaluated with a Bonferroni correction.Results: Mixed‐effects modeling showed faster volumetric declines in the caudal middle frontal cortex, fusiform, inferior frontal, superior temporal (STG), and supramarginal (SMG) gyri, at −0.046 to −0.138 cm3/year in individuals with prior‐year alcohol and cannabis co‐use, but not those engaged in alcohol or cannabis use only. These findings cannot be explained by more severe alcohol use among co‐users. Further, alcohol and cannabis co‐use in early versus late adolescence predicted faster volumetric decline in the STG and SMG across assigned sex at birth.Conclusions: Findings highlight the longitudinal impact of alcohol and cannabis co‐use on brain development, especially among youth reporting early adolescent onset of use. The volumetric decline was noted in cortical regions in support of attention, memory, executive control, and social cognition, suggesting the pervasive effect of alcohol and cannabis co‐use on brain development.
- 12 July 2022
- Published by Wiley in Addiction Biology
- Vol. 27 (5) , e13208
Alcohol and cannabis co‐use and longitudinal gray matter volumetric changes in early and late adolescence
Background: Previous studies have characterized the impact of substance use on cerebral structure and function in adolescents. Yet, the great majority of prior studies employed a small sample, presented cross‐sectional findings, and omitted potential sex differences.Methods: Using data based on 724 adolescents (370 females) curated from the NCANDA study, we investigated how gray matter volumes (GMVs) decline longitudinally as a result of alcohol and cannabis use. The impacts of alcohol and cannabis co‐use and how these vary across assigned sex at birth and age were examined. Brain imaging data comprised the GMVs of 34 regions of interest and the results were evaluated with a Bonferroni correction.Results: Mixed‐effects modeling showed faster volumetric declines in the caudal middle frontal cortex, fusiform, inferior frontal, superior temporal (STG), and supramarginal (SMG) gyri, at −0.046 to −0.138 cm3/year in individuals with prior‐year alcohol and cannabis co‐use, but not those engaged in alcohol or cannabis use only. These findings cannot be explained by more severe alcohol use among co‐users. Further, alcohol and cannabis co‐use in early versus late adolescence predicted faster volumetric decline in the STG and SMG across assigned sex at birth.Conclusions: Findings highlight the longitudinal impact of alcohol and cannabis co‐use on brain development, especially among youth reporting early adolescent onset of use. The volumetric decline was noted in cortical regions in support of attention, memory, executive control, and social cognition, suggesting the pervasive effect of alcohol and cannabis co‐use on brain development.
- Open Access
- 10 January 2022
- Published by Springer Nature in BMC Bioinformatics
- Vol. 23 (1) , 1-15
Polygenic risk prediction based on singular value decomposition with applications to alcohol use disorder
Background/aim: The polygenic risk score (PRS) shows promise as a potentially effective approach to summarize genetic risk for complex diseases such as alcohol use disorder that is influenced by a combination of multiple variants, each of which has a very small effect. Yet, conventional PRS methods tend to over-adjust confounding factors in the discovery sample and thus have low power to predict the phenotype in the target sample. This study aims to address this important methodological issue.Methods: This study proposed a new method to construct PRS by (1) approximating the polygenic model using a few principal components selected based on eigen-correlation in the discovery data; and (2) conducting principal component projection on the target data. Secondary data analysis was conducted on two large scale databases: the Study of Addiction: Genetics and Environment (SAGE; discovery data) and the National Longitudinal Study of Adolescent to Adult Health (Add Health; target data) to compare performance of the conventional and proposed methods.Result and conclusion: The results show that the proposed method has higher prediction power and can handle participants from different ancestry backgrounds. We also provide practical recommendations for setting the linkage disequilibrium (LD) andpvalue thresholds.
- 10 January 2022
- Published by Springer Nature in BMC Bioinformatics
- Vol. 23 (1) , 1-15
Open AccessPolygenic risk prediction based on singular value decomposition with applications to alcohol use disorder
Background/aim: The polygenic risk score (PRS) shows promise as a potentially effective approach to summarize genetic risk for complex diseases such as alcohol use disorder that is influenced by a combination of multiple variants, each of which has a very small effect. Yet, conventional PRS methods tend to over-adjust confounding factors in the discovery sample and thus have low power to predict the phenotype in the target sample. This study aims to address this important methodological issue.Methods: This study proposed a new method to construct PRS by (1) approximating the polygenic model using a few principal components selected based on eigen-correlation in the discovery data; and (2) conducting principal component projection on the target data. Secondary data analysis was conducted on two large scale databases: the Study of Addiction: Genetics and Environment (SAGE; discovery data) and the National Longitudinal Study of Adolescent to Adult Health (Add Health; target data) to compare performance of the conventional and proposed methods.Result and conclusion: The results show that the proposed method has higher prediction power and can handle participants from different ancestry backgrounds. We also provide practical recommendations for setting the linkage disequilibrium (LD) andpvalue thresholds.
- Open Access
- 17 February 2020
- Published by Institute of Mathematical Statistics in The Annals of Statistics
- Vol. 48 (1) , 111-137
Model assisted variable clustering: Minimax-optimal recovery and algorithms
The problem of variable clustering is that of estimating groups of similar components of a $p$-dimensional vector $X=(X_{1},\ldots ,X_{p})$ from $n$ independent copies of $X$. There exists a large number of algorithms that return data-dependent groups of variables, but their interpretation is limited to the algorithm that produced them. An alternative is model-based clustering, in which one begins by defining population level clusters relative to a model that embeds notions of similarity. Algorithms tailored to such models yield estimated clusters with a clear statistical interpretation. We take this view here and introduce the class of $G$-block covariance models as a background model for variable clustering. In such models, two variables in a cluster are deemed similar if they have similar associations will all other variables. This can arise, for instance, when groups of variables are noise corrupted versions of the same latent factor. We quantify the difficulty of clustering data generated from a $G$-block covariance model in terms of cluster proximity, measured with respect to two related, but different, cluster separation metrics. We derive minimax cluster separation thresholds, which are the metric values below which no algorithm can recover the model-defined clusters exactly, and show that they are different for the two metrics. We therefore develop two algorithms, COD and PECOK, tailored to $G$-block covariance models, and study their minimax-optimality with respect to each metric. Of independent interest is the fact that the analysis of the PECOK algorithm, which is based on a corrected convex relaxation of the popular $K$-means algorithm, provides the first statistical analysis of such algorithms for variable clustering. Additionally, we compare our methods with another popular clustering method, spectral clustering. Extensive simulation studies, as well as our data analyses, confirm the applicability of our approach.- 17 February 2020
- Published by Institute of Mathematical Statistics in The Annals of Statistics
- Vol. 48 (1) , 111-137
Open AccessModel assisted variable clustering: Minimax-optimal recovery and algorithms
The problem of variable clustering is that of estimating groups of similar components of a $p$-dimensional vector $X=(X_{1},\ldots ,X_{p})$ from $n$ independent copies of $X$. There exists a large number of algorithms that return data-dependent groups of variables, but their interpretation is limited to the algorithm that produced them. An alternative is model-based clustering, in which one begins by defining population level clusters relative to a model that embeds notions of similarity. Algorithms tailored to such models yield estimated clusters with a clear statistical interpretation. We take this view here and introduce the class of $G$-block covariance models as a background model for variable clustering. In such models, two variables in a cluster are deemed similar if they have similar associations will all other variables. This can arise, for instance, when groups of variables are noise corrupted versions of the same latent factor. We quantify the difficulty of clustering data generated from a $G$-block covariance model in terms of cluster proximity, measured with respect to two related, but different, cluster separation metrics. We derive minimax cluster separation thresholds, which are the metric values below which no algorithm can recover the model-defined clusters exactly, and show that they are different for the two metrics. We therefore develop two algorithms, COD and PECOK, tailored to $G$-block covariance models, and study their minimax-optimality with respect to each metric. Of independent interest is the fact that the analysis of the PECOK algorithm, which is based on a corrected convex relaxation of the popular $K$-means algorithm, provides the first statistical analysis of such algorithms for variable clustering. Additionally, we compare our methods with another popular clustering method, spectral clustering. Extensive simulation studies, as well as our data analyses, confirm the applicability of our approach. - Open Access
Time-dependent Canonical Correlation Analysis for Multilevel Time Series
Canonical Correlation Analysis is a technique in multivariate data analysis for finding linear projections that maximize the correlation between two groups of variables. The correlations are typically defined without accounting for the serial correlations between observations, a typical setting for time series data. To understand the coupling dynamics and temporal variations between the two time-varying sources, we introduce the time-dependent canonical correlation analysis (TDCCA), a method for inferring time-dependent canonical vectors from multilevel time series data. A convex formulation of the problem is proposed, which leverages the singular value decomposition (SVD) characterization of all solutions of the CCA problem. We use simulated datasets to validate the proposed algorithm. Moreover, we propose a novel measure, canonical correlation variation as another way to assess the dynamic pattern of brain connections and we apply it to a real resting state fMRI dataset to study the aging effects on brain connectivity. Additionally, we explore our proposed method in a task-related fMRI to detect the temporal dynamics due to different motor tasks. We show that, compared to extant methods, the TDCCA-based approach not only detect temporal changes but also improves feature extraction. Together, this paper contributes broadly to new computational methodologies in understanding multilevel time series.
Open AccessTime-dependent Canonical Correlation Analysis for Multilevel Time Series
Canonical Correlation Analysis is a technique in multivariate data analysis for finding linear projections that maximize the correlation between two groups of variables. The correlations are typically defined without accounting for the serial correlations between observations, a typical setting for time series data. To understand the coupling dynamics and temporal variations between the two time-varying sources, we introduce the time-dependent canonical correlation analysis (TDCCA), a method for inferring time-dependent canonical vectors from multilevel time series data. A convex formulation of the problem is proposed, which leverages the singular value decomposition (SVD) characterization of all solutions of the CCA problem. We use simulated datasets to validate the proposed algorithm. Moreover, we propose a novel measure, canonical correlation variation as another way to assess the dynamic pattern of brain connections and we apply it to a real resting state fMRI dataset to study the aging effects on brain connectivity. Additionally, we explore our proposed method in a task-related fMRI to detect the temporal dynamics due to different motor tasks. We show that, compared to extant methods, the TDCCA-based approach not only detect temporal changes but also improves feature extraction. Together, this paper contributes broadly to new computational methodologies in understanding multilevel time series.
- 22 April 2019
- Published by Oxford University Press (OUP) in Biometrics
- Vol. 75 (3) , 788-798
Granger Mediation Analysis of Multiple Time Series With an Application to Functional Magnetic Resonance Imaging
This paper presents Granger mediation analysis, a new framework for causal mediation analysis of multiple time series. This framework is motivated by a functional magnetic resonance imaging (fMRI) experiment where we are interested in estimating the mediation effects between a randomized stimulus time series and brain activity time series from two brain regions. The independent observation assumption is thus unrealistic for this type of time-series data. To address this challenge, our framework integrates two types of models: causal mediation analysis across the mediation variables, and vector autoregressive (VAR) models across the temporal observations. We use “Granger” to refer to VAR correlations modeled in this paper. We further extend this framework to handle multilevel data, in order to model individual variability and correlated errors between the mediator and the outcome variables. Using Rubin's potential outcome framework, we show that the causal mediation effects are identifiable under our time-series model. We further develop computationally efficient algorithms to maximize our likelihood-based estimation criteria. Simulation studies show that our method reduces the estimation bias and improves statistical power, compared with existing approaches. On a real fMRI data set, our approach quantifies the causal effects through a brain pathway, while capturing the dynamic dependence between two brain regions.
- 22 April 2019
- Published by Oxford University Press (OUP) in Biometrics
- Vol. 75 (3) , 788-798
Granger Mediation Analysis of Multiple Time Series With an Application to Functional Magnetic Resonance Imaging
This paper presents Granger mediation analysis, a new framework for causal mediation analysis of multiple time series. This framework is motivated by a functional magnetic resonance imaging (fMRI) experiment where we are interested in estimating the mediation effects between a randomized stimulus time series and brain activity time series from two brain regions. The independent observation assumption is thus unrealistic for this type of time-series data. To address this challenge, our framework integrates two types of models: causal mediation analysis across the mediation variables, and vector autoregressive (VAR) models across the temporal observations. We use “Granger” to refer to VAR correlations modeled in this paper. We further extend this framework to handle multilevel data, in order to model individual variability and correlated errors between the mediator and the outcome variables. Using Rubin's potential outcome framework, we show that the causal mediation effects are identifiable under our time-series model. We further develop computationally efficient algorithms to maximize our likelihood-based estimation criteria. Simulation studies show that our method reduces the estimation bias and improves statistical power, compared with existing approaches. On a real fMRI data set, our approach quantifies the causal effects through a brain pathway, while capturing the dynamic dependence between two brain regions.
- Open Access
Model Assisted Variable Clustering: Minimax-optimal Recovery and Algorithms
Model-based clustering defines population level clusters relative to a model that embeds notions of similarity. Algorithms tailored to such models yield estimated clusters with a clear statistical interpretation. We take this view here and introduce the class of G-block covariance models as a background model for variable clustering. In such models, two variables in a cluster are deemed similar if they have similar associations will all other variables. This can arise, for instance, when groups of variables are noise corrupted versions of the same latent factor. We quantify the difficulty of clustering data generated from a G-block covariance model in terms of cluster proximity, measured with respect to two related, but different, cluster separation metrics. We derive minimax cluster separation thresholds, which are the metric values below which no algorithm can recover the model-defined clusters exactly, and show that they are different for the two metrics. We therefore develop two algorithms, COD and PECOK, tailored to G-block covariance models, and study their minimax-optimality with respect to each metric. Of independent interest is the fact that the analysis of the PECOK algorithm, which is based on a corrected convex relaxation of the popular K-means algorithm, provides the first statistical analysis of such algorithms for variable clustering. Additionally, we contrast our methods with another popular clustering method, spectral clustering, specialized to variable clustering, and show that ensuring exact cluster recovery via this method requires clusters to have a higher separation, relative to the minimax threshold. Extensive simulation studies, as well as our data analyses, confirm the applicability of our approach.
Open AccessModel Assisted Variable Clustering: Minimax-optimal Recovery and Algorithms
Model-based clustering defines population level clusters relative to a model that embeds notions of similarity. Algorithms tailored to such models yield estimated clusters with a clear statistical interpretation. We take this view here and introduce the class of G-block covariance models as a background model for variable clustering. In such models, two variables in a cluster are deemed similar if they have similar associations will all other variables. This can arise, for instance, when groups of variables are noise corrupted versions of the same latent factor. We quantify the difficulty of clustering data generated from a G-block covariance model in terms of cluster proximity, measured with respect to two related, but different, cluster separation metrics. We derive minimax cluster separation thresholds, which are the metric values below which no algorithm can recover the model-defined clusters exactly, and show that they are different for the two metrics. We therefore develop two algorithms, COD and PECOK, tailored to G-block covariance models, and study their minimax-optimality with respect to each metric. Of independent interest is the fact that the analysis of the PECOK algorithm, which is based on a corrected convex relaxation of the popular K-means algorithm, provides the first statistical analysis of such algorithms for variable clustering. Additionally, we contrast our methods with another popular clustering method, spectral clustering, specialized to variable clustering, and show that ensuring exact cluster recovery via this method requires clusters to have a higher separation, relative to the minimax threshold. Extensive simulation studies, as well as our data analyses, confirm the applicability of our approach.
- Open Access
Covariate Assisted Principal Regression for Covariance Matrix Outcomes
Modeling variances in data has been an important topic in many fields, including in financial and neuroimaging analysis. We consider the problem of regressing covariance matrices on a vector covariates, collected from each observational unit. The main aim is to uncover the variation in the covariance matrices across units that are explained by the covariates. This paper introduces Covariate Assisted Principal (CAP) regression, an optimization-based method for identifying the components predicted by (generalized) linear models of the covariates. We develop computationally efficient algorithms to jointly search the projection directions and regression coefficients, and we establish the asymptotic properties. Using extensive simulation studies, our method shows higher accuracy and robustness in coefficient estimation than competing methods. Applied to a resting-state functional magnetic resonance imaging study, our approach identifies the human brain network changes associated with age and sex.
Open AccessCovariate Assisted Principal Regression for Covariance Matrix Outcomes
Modeling variances in data has been an important topic in many fields, including in financial and neuroimaging analysis. We consider the problem of regressing covariance matrices on a vector covariates, collected from each observational unit. The main aim is to uncover the variation in the covariance matrices across units that are explained by the covariates. This paper introduces Covariate Assisted Principal (CAP) regression, an optimization-based method for identifying the components predicted by (generalized) linear models of the covariates. We develop computationally efficient algorithms to jointly search the projection directions and regression coefficients, and we establish the asymptotic properties. Using extensive simulation studies, our method shows higher accuracy and robustness in coefficient estimation than competing methods. Applied to a resting-state functional magnetic resonance imaging study, our approach identifies the human brain network changes associated with age and sex.