Grant: Discovery Grants Program - Individual
Institut national de la recherche scientifique
$27,000.00
Multivariate Analysis and High-Dimension Dependent Genomics Modelling using Expectile
EF 2024-25
Résumé
Recent technological advances have enabled the generation of large amounts of genomic and neuroimaging data, which raises considerable methodological and computational challenges. In practice, these complex, heterogeneous and often error-prone data are analyzed with statistical methods based on the mean (MBM). However, it is well known that MBM rely on oversimplified assumptions and are inadequate to model the data complexity. To address these challenges, my long-term research program involves the development and integration of expectile-based methods (EBM) to classical statistical inference and prediction, and its application to genomic and neuroimaging data, and the application of m-quantiles to address robustness.
Expectiles are weighted averages that generalize the mean and describe the random variable distribution function, as do quantiles. EBM generalize classical MBM by modelling the expectiles of the conditional response distribution. They enable to capture the unobserved heterogeneity present in the data and allow a detailed statistical analysis of the covariate effects on the entire response distribution, without specifying any assumptions about the error distribution. EBM have received increased interest over the past decade. However, their application is not well established among applied statisticians and practitioners who continue to rely on MBM.
Classical inference methods such as analysis of variance (ANOVA), Wald test or Goodness-of-fit test, to name a few, remain standard methods frequently used in applied statistics. However, all these inference methods focus on the mean, which is inefficient. The first short-to-mid-term goal of my research is to adapt EBM to these classical inference methods.
High-dimension family genomic data are very important in identifying causal variants. However, besides being high-dimensional, they are dependent and multigenerational. The second short-to-mid-term goal of my research focuses on integrating EBM to high-dimension dependent genomic data. This research avenue will generalize the MBM and capture more variability and heteroscedasticity in complex data with non-normal, multimodal and asymmetric error distributions.
High-dimension genomic data are also prone to noise during acquisition or pre-processing steps and to missing data. My third medium-term goal is to develop high-dimension EBM for dependent genomic data which cope with measurement error and missing data challenges.
EBM can play a key role in many health-related research areas. In our work, we applied EBM to MRI images to improve the accuracy of age prediction which is a score of various neurodegenerative diseases. My long-term research program will enable identification of new causal variants in the genome-wise association studies, which is an MBM. This EBM research will be leveraged to build an efficient and powerful algorithm to boost the predictive power of the polygenic risk score in populations of non-European ancestry.
Programme
Discovery Grants Program - Individual
Bénéficiaire
Emplacement
Quebec, Canada
Type de bénéficiaire
Academia
Type
Grant
Source
NSERC's Awards Data
Tel que publié
Application id
RGPIN-2023-05238
Name nom
Barry, Amadou
Department département
Armand-Frappier Santé Biotechnologie
Organization id
56
Institution établissement
Institut national de la recherche scientifique
Province en
Québec
Province fr
Québec
Country en
CANADA
Country fr
CANADA
Fiscal year exercice financier
2024
Competition year année de concours
2023
Award amount
27000
Program id
RGPIN
Program name en
Discovery Grants Program - Individual
Program name fr
Programme de subventions à la découverte - individuelles
Group en
DISCOVERY RESEARCH
Group fr
RECHERCHE AXÉE SUR LA DÉCOUVERTE
Committee code
1508
Committee name en
Mathematics and Statistics
Committee name fr
Mathématiques et statistique
Area of application code
1208
Area of application group en
Advancement of knowledge
Area of application group fr
Avancement des connaissances
Area of application en
Materials sciences
Area of application fr
Sciences des matériaux
Research subject code
3011
Research subject group en
Statistics and probability
Research subject group fr
Statistique et probabilité
Research subject en
Multivariate analysis
Research subject fr
Analyse multivariée
Field of research list names en
Not available
Field of research list names fr
Non disponible
Socioeconomic objective list names en
Not available
Socioeconomic objective list names fr
Non disponible
Application title
Multivariate Analysis and High-Dimension Dependent Genomics Modelling using Expectile
Keywords
expectile, m-quantile, and quantile; inference and modelisation; statistical genomics and neuroscience; high-dimension; dependent modelling; multivariate analysis; measurement error; longitudinal analysis; missing data; causal inference
Application summary
Recent technological advances have enabled the generation of large amounts of genomic and neuroimaging data, which raises considerable methodological and computational challenges. In practice, these complex, heterogeneous and often error-prone data are analyzed with statistical methods based on the mean (MBM). However, it is well known that MBM rely on oversimplified assumptions and are inadequate to model the data complexity. To address these challenges, my long-term research program involves the development and integration of expectile-based methods (EBM) to classical statistical inference and prediction, and its application to genomic and neuroimaging data, and the application of m-quantiles to address robustness.
Expectiles are weighted averages that generalize the mean and describe the random variable distribution function, as do quantiles. EBM generalize classical MBM by modelling the expectiles of the conditional response distribution. They enable to capture the unobserved heterogeneity present in the data and allow a detailed statistical analysis of the covariate effects on the entire response distribution, without specifying any assumptions about the error distribution. EBM have received increased interest over the past decade. However, their application is not well established among applied statisticians and practitioners who continue to rely on MBM.
Classical inference methods such as analysis of variance (ANOVA), Wald test or Goodness-of-fit test, to name a few, remain standard methods frequently used in applied statistics. However, all these inference methods focus on the mean, which is inefficient. The first short-to-mid-term goal of my research is to adapt EBM to these classical inference methods.
High-dimension family genomic data are very important in identifying causal variants. However, besides being high-dimensional, they are dependent and multigenerational. The second short-to-mid-term goal of my research focuses on integrating EBM to high-dimension dependent genomic data. This research avenue will generalize the MBM and capture more variability and heteroscedasticity in complex data with non-normal, multimodal and asymmetric error distributions.
High-dimension genomic data are also prone to noise during acquisition or pre-processing steps and to missing data. My third medium-term goal is to develop high-dimension EBM for dependent genomic data which cope with measurement error and missing data challenges.
EBM can play a key role in many health-related research areas. In our work, we applied EBM to MRI images to improve the accuracy of age prediction which is a score of various neurodegenerative diseases. My long-term research program will enable identification of new causal variants in the genome-wise association studies, which is an MBM. This EBM research will be leveraged to build an efficient and powerful algorithm to boost the predictive power of the polygenic risk score in populations of non-European ancestry.