Novel cell state discovery from higher-order interactions in single cell expression data

Objectives:



The aim of this project is to investigate higher-order gene interactions in single cell RNA sequencing data and examine how these interaction networks define novel cell states across developmental or disease trajectories. In particular, we aim to quantify how a set of transcription factors and their target genes interact differently under various biological conditions leading to different cell states.



Novel Science:



Current approaches of investigating complex biological systems often rely on using machine learning to fit parametric models, with a major drawback being that they potentially miss the ground truth, if the restrictions of the parametric fits are not aligned with the real data-generating biological process. This project will utilize novel statistical techniques for the non-parametric, multi-order estimation of symmetric interactions developed by A. Khamseh and S. Beentjes, motivated by the Targeted Learning paradigm originated by Mark Van Der Laan. The project will expand on the interaction analysis pipeline by A. Jansma, which has already shown promising results in identifying both cell types and states in an unsupervised manner, with clustering based on gene expression dependency networks. The methods generated throughout this project will expand on this work to give a more comprehensive and deeper description of mechanisms present in complex biological systems, which allows for quantitative and biologically-interpretable comparison of cellular systems at different time-points, and/or states.