University of British Columbia
Nonparametric estimation in high dimensions with applications to transcriptomics data
Abstract
dc:descriptionThe emergence of massive, high-dimensional datasets in single-cell biology has invited the development of novel statistical methodology and analyses. The complexity of transcriptomics data warrants the use of nonparametric methods, which provide superior modelling flexibility and expressiveness. However, nonparametric methods are prone to suffer from statistical and computational challenges in high dimensions, with significant implications for algorithm and experiment design. This thesis presents theoretical and computational advances addressing high-dimensional problems typical of transcriptomics, namely, denoising, sampling design and density estimation. First, we present an approach to denoising spatial transcriptomics images that is particularly effective for uncovering cell identities in the regime of ultra-low sequencing depths, and also allows for interpolation of gene expression. The method – Spatial Transcriptomics via Adaptive Regularization and Kernels (STARK) – augments kernel ridge regression with an incrementally adaptive graph Laplacian regularizer. Starting with a purely spatial graph and updating it as we improve our image estimate makes the method more robust to noise in low-sequencing-depth regimes. Second, we analyze noisy empirical distributions of cells in single-cell transcriptomics, obtaining upper and lower bounds on the Wasserstein distance to the true population. These bounds prove useful in determining the optimal number of cells to sample given a finite budget of sequencing reads, or equivalently, in determining the optimal sequencing depth. Third, we propose a method for estimating log-concave densities on \mathbb Rd under an independent components assumption. While multivariate log-concave density estimation is hard both computationally and statistically, we show that the independent components assumption alleviates both issues, while still maintaining a large nonparametric class.
Degree
thesis:*- Name thesis:degree_name
- Doctor of Philosophy - PhD
- Level thesis:degree_level
- doctoral
- Discipline thesis:degree_discipline
- Mathematics
- Grantor dc:publisher
- University of British Columbia
- Year dc:date
- 2026
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Kubal, Sharvaj
- Contributors dc:contributor
-
- Schiebinger, Geoffrey
- Plan, Yaniv
Rights
dc:rights- Statement dc:rights
-
- Attribution-NonCommercial-NoDerivatives 4.0 International
- Language dc:language
- eng
Identifiers
dc:identifier.*- Handle dc:identifier
- http://hdl.handle.net/2429/95209