Canada Spends Logo

Explore our Budget 2026 analysis

Grant: Discovery Grants Program - Individual

University of Toronto

$27,000.00

Random matrices with low-rank structure for inferring chromatin conformation architecture

FY 2024-25

Summary
Low-rank matrix approximation (LRMA) arises in a variety of core statistical and machine learning techniques including dimension reduction, denoising, and data compression, and is central to many multivariate statistical techniques such as principal components, multidimensional scaling, linear discriminant analysis, and canonical correlation analysis. The main goal of the LRMA technique is to identify the "best" way to approximate some given matrix with a low-rank version, where the "goodness of fit" between a matrix and its approximation is typically measured in the form of a residual sum of squares (partial or weighted for some tasks). Due to the convenient link between Gaussian log-likelihood and residual sums of squares, in most cases, the probabilistic formulation of the low-rank matrix approximation problem is limited to the Gaussian distribution. The first long-term objective of this research program will be to develop a general probabilistic approach to modeling matrices with a low-rank structure, inspired by previous results of the applicant in the chromatin conformation field. The general approach makes two assumptions about the input data matrix: (1) the matrix elements are generated from the same exponential family distribution, and (2) there is a small set of common latent factors that influence the distribution of each element. In the model, these factors are assumed to be unknown, resulting in a low-rank structure incorporated into the parameters of the distribution. There are many practical applications for random matrices with a low-rank structure. In ecology, Poisson low-rank models can be especially useful when modeling population matrices (with rows and columns corresponding to sites and species, respectively). Bernoulli low-rank models can find applications in item response theory, where the input matrix is binary and indicates whether a particular individual will answer a question correctly. The second long-term objective of my research program will focus on advancing the field of computational biology and using the proposed methodology for a particular application to chromatin conformation reconstruction. The three-dimensional structure of chromatin (a complex of DNA and proteins comprising chromosomes) plays a crucial role in many vital cellular processes, which makes conformation reconstruction one of the most crucial tasks in modern computational biology. The central object in this study is the so-called contact matrix, which represents how often pairs of chromatin pieces are "in contact", i.e. adjacent in 3D space. Because of the apparent link between contact counts and pairwise distances between parts of chromatin as well as the embedding in three-dimensional space that naturally leads to a low-rank constraint, contact matrices can be considered as a special case of random matrices with a low-rank structure.
Program
Discovery Grants Program - Individual
Location
Ontario, Canada
Recipient Type
Academia
Type
Grant
Source
NSERC's Awards Data

As published

Application id
RGPIN-2023-04727
Name nom
Tuzhilina, Elena
Department département
Statistical Sciences
Organization id
31
Institution établissement
University of Toronto
Province en
Ontario
Province fr
Ontario
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
1102
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
Biomedical engineering
Area of application fr
Génie biomédical
Research subject code
3004
Research subject group en
Statistics and probability
Research subject group fr
Statistique et probabilité
Research subject en
Applied statistics
Research subject fr
Statistique appliquée
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
Random matrices with low-rank structure for inferring chromatin conformation architecture
Keywords
statistical modeling; dimension reduction; smoothing splines; multi-dimensional scaling; computational biology; genome reconstruction; 3d spatial structure; pairwise proximity data
Application summary
Low-rank matrix approximation (LRMA) arises in a variety of core statistical and machine learning techniques including dimension reduction, denoising, and data compression, and is central to many multivariate statistical techniques such as principal components, multidimensional scaling, linear discriminant analysis, and canonical correlation analysis. The main goal of the LRMA technique is to identify the "best" way to approximate some given matrix with a low-rank version, where the "goodness of fit" between a matrix and its approximation is typically measured in the form of a residual sum of squares (partial or weighted for some tasks). Due to the convenient link between Gaussian log-likelihood and residual sums of squares, in most cases, the probabilistic formulation of the low-rank matrix approximation problem is limited to the Gaussian distribution. The first long-term objective of this research program will be to develop a general probabilistic approach to modeling matrices with a low-rank structure, inspired by previous results of the applicant in the chromatin conformation field. The general approach makes two assumptions about the input data matrix: (1) the matrix elements are generated from the same exponential family distribution, and (2) there is a small set of common latent factors that influence the distribution of each element. In the model, these factors are assumed to be unknown, resulting in a low-rank structure incorporated into the parameters of the distribution. There are many practical applications for random matrices with a low-rank structure. In ecology, Poisson low-rank models can be especially useful when modeling population matrices (with rows and columns corresponding to sites and species, respectively). Bernoulli low-rank models can find applications in item response theory, where the input matrix is binary and indicates whether a particular individual will answer a question correctly. The second long-term objective of my research program will focus on advancing the field of computational biology and using the proposed methodology for a particular application to chromatin conformation reconstruction. The three-dimensional structure of chromatin (a complex of DNA and proteins comprising chromosomes) plays a crucial role in many vital cellular processes, which makes conformation reconstruction one of the most crucial tasks in modern computational biology. The central object in this study is the so-called contact matrix, which represents how often pairs of chromatin pieces are "in contact", i.e. adjacent in 3D space. Because of the apparent link between contact counts and pairwise distances between parts of chromatin as well as the embedding in three-dimensional space that naturally leads to a low-rank constraint, contact matrices can be considered as a special case of random matrices with a low-rank structure.