Designing new multimodal data analysis approaches by geometry, topology and data diffusion

Our goal is to uncover patterns and structures hidden in complex biological data measured at the single-cell resolution. We propose to incorporate multiscale information, comprising both local and global information, to describe the shape of datasets as a whole. Current work typically only focuses on local structures, like groups of similar cells and transitions between them, missing out on describing the underlying organizational principles of data. Here, we will develop quantitative methods by making use of our recent work on diffusion condensation, a process that summarizes data at different resolutions that go from very fine-grained to very coarse-grained. We will tackle challenging tasks in single-cell data analysis, such as removing biases like batch effects from the data collection process and combining multiple modalities, i.e., datasets measured with different instruments. Our methods will incorporate information at multiple scales of locality, ranging from local to global, thus enabling the robust identification of patterns in the data that (a) represent the underlying biology and, (b) are shared across datasets. This research bears the promise of leading to novel insights of complex disease mechanisms, helping domain scientists like biologists to make sense of their data and assist them in hypothesis generation.

Faculty Supervisor:

Guy Wolf

Student:

Partner:

Helmholtz Centre Munich

Discipline:

Computer science

Sector:

Artificial Intelligence

University:

Université de Montréal

Program: