Canada Spends Logo

Explorez notre analyse du budget 2026

Grant: Discovery Launch Supplement

University of Alberta

$12,500.00

Transfer learning for imbalanced data

EF 2024-25

Résumé
Individuals with rare diseases need more support.  Collaboration between experts, researchers, and clinicians is essential to improve treatments and cures for those with rare diseases. However, the rarity and diversity of these diseases present challenges. There are many more healthy individuals than those with rare diseases, creating imbalanced datasets. This imbalance causes problems for algorithms, which can be biased toward the majority class and perform poorly on the minority class. The minority class (those with rare diseases) is often more important. Case-control studies are crucial for investigating factors linked to rare diseases. They sample equal numbers of diseased cases and healthy controls, which can be used to explore potential risk factors for a disease among cases versus controls. However, they may fail to reveal true disease proportions or predict well due to biased intercept estimates.   In this study, we aim to address the issue of imbalanced data by leveraging transfer learning techniques. Transfer learning has demonstrated its effectiveness in transferring information from related problems with more balanced population proportions. In many cases, the objective is to learn the exposure association of an outcome for a specific target population, while the available data may come from various other sources that share certain similarities with the target. For instance, the Illumina HumanMethylation450 arrays comprise 526 experiments with distinct purposes collected through case-control designs, including trials on Type 1 diabetes (E-GEOD-54564) and colorectal cancer (EMTAB-8148). Although the cases differ, the controls are non-disease samples randomly collected from the population, and all accessions collect the transcript profiles from the same array (covariates). Therefore, we propose to apply transfer learning to the target case-control study (Type 1 diabetes) and the source case-control study (colorectal cancer) based on the same gene array but with different purposes. The primary goal is to transfer the information from the sources to the target to improve the estimation and inference under case-control studies, which helps the model adapt to heterogeneity in feature distributions and model coefficients. We intend to propose a two-step joint estimator for the target case-control studies. The first step aims to obtain a joint estimator of the coefficients by maximizing log-likelihood using the target data. In the second step, we will leverage the joint estimator to obtain a debiased estimator using the source data. We will present the theoretical properties of the proposed two-step estimator, including the convergence rate, asymptotic consistency, and normality. Furthermore, we will implement the proposed method in numerical studies, including rare disease data. In summary, the proposal aims to develop better statistical tools for studying imbalanced case-control data across populations.
Programme
Discovery Launch Supplement
Bénéficiaire
Emplacement
Alberta, Canada
Type de bénéficiaire
Academia
Type
Grant
Source
NSERC's Awards Data

Tel que publié

Application id
DGECR-2024-00333
Name nom
TANG, Wenlu
Department département
Department of Mathematical and Statistical Sciences
Organization id
9
Institution établissement
University of Alberta
Province en
Alberta
Province fr
Alberta
Country en
CANADA
Country fr
CANADA
Fiscal year exercice financier
2024
Competition year année de concours
2024
Award amount
12500
Program id
DGECR
Program name en
Discovery Launch Supplement
Program name fr
Supplément Tremplin vers la découverte
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
1101
Area of application group en
Health, education and social services
Area of application group fr
Santé publique, éducation et services sociaux
Area of application en
Human health (including medically-related psychological research)
Area of application fr
Santé publique (y compris recherche psychologique liée à la médecine)
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
Transfer learning for imbalanced data
Keywords
transfer learning; imbalanced data; case-control studies; maximum likelihood estimation; biostatistics
Application summary
Individuals with rare diseases need more support.  Collaboration between experts, researchers, and clinicians is essential to improve treatments and cures for those with rare diseases. However, the rarity and diversity of these diseases present challenges. There are many more healthy individuals than those with rare diseases, creating imbalanced datasets. This imbalance causes problems for algorithms, which can be biased toward the majority class and perform poorly on the minority class. The minority class (those with rare diseases) is often more important. Case-control studies are crucial for investigating factors linked to rare diseases. They sample equal numbers of diseased cases and healthy controls, which can be used to explore potential risk factors for a disease among cases versus controls. However, they may fail to reveal true disease proportions or predict well due to biased intercept estimates.   In this study, we aim to address the issue of imbalanced data by leveraging transfer learning techniques. Transfer learning has demonstrated its effectiveness in transferring information from related problems with more balanced population proportions. In many cases, the objective is to learn the exposure association of an outcome for a specific target population, while the available data may come from various other sources that share certain similarities with the target. For instance, the Illumina HumanMethylation450 arrays comprise 526 experiments with distinct purposes collected through case-control designs, including trials on Type 1 diabetes (E-GEOD-54564) and colorectal cancer (EMTAB-8148). Although the cases differ, the controls are non-disease samples randomly collected from the population, and all accessions collect the transcript profiles from the same array (covariates). Therefore, we propose to apply transfer learning to the target case-control study (Type 1 diabetes) and the source case-control study (colorectal cancer) based on the same gene array but with different purposes. The primary goal is to transfer the information from the sources to the target to improve the estimation and inference under case-control studies, which helps the model adapt to heterogeneity in feature distributions and model coefficients. We intend to propose a two-step joint estimator for the target case-control studies. The first step aims to obtain a joint estimator of the coefficients by maximizing log-likelihood using the target data. In the second step, we will leverage the joint estimator to obtain a debiased estimator using the source data. We will present the theoretical properties of the proposed two-step estimator, including the convergence rate, asymptotic consistency, and normality. Furthermore, we will implement the proposed method in numerical studies, including rare disease data. In summary, the proposal aims to develop better statistical tools for studying imbalanced case-control data across populations.