Manifold Fitting and Its Potential Applications in the Analysis of Sequencing Data
组织者
演讲者
卢昱堃
时间
2024年11月11日 09:00 至 09:30
地点
Online
摘要
In recent years, manifold fitting has emerged as a key technique in non-Euclidean statistical analysis, aimed at recovering low-dimensional structures underlying high-dimensional data. Traditional methods approximate these manifolds by estimating tangent spaces at each data point, but they often assume bounded noise, which limits accuracy when noise is unbounded. Our approach addresses this by estimating tangent spaces directly at projected points on the manifold, thus enhancing manifold fitting accuracy in noisy conditions. This foundational work has inspired further applications, including scAMF (Single-cell Analysis via Manifold Fitting), which improves clustering and visualization in single-cell RNA sequencing (scRNA-seq) by reducing noise and aligning gene expression vectors with their true structures. Ongoing efforts are expanding this methodology to spatial transcriptomic data transformation and isoform detection in long-read sequencing data, aiming to refine data representations in these complex biological contexts. These advancements underline the potential of manifold fitting techniques in driving progress across high-dimensional biological data analysis.