Computational Strategies for Deciphering Cellular Differentiation and Cancer Progression

Fred Hutchinson Cancer Center, Seattle WA

Investigators

Abstract

Project Summary Many diseases arise from molecular alterations within cellular mechanisms, and our understanding of these processes is limited. Single-cell measurement technologies provide rich molecular data but current computational methods struggle with their complexity. Traditional approaches treat single-cell data as discrete sets, obscuring subtle continuous biological phenomena. To address this, we developed Mellon, an algorithm that offers a continuous representation of high-dimensional single-cell landscapes by inferring a continuous density function, effectively characterizing the distribution of cell states in a biological system. This proposal aims to develop robust tools for modeling and interpreting single-cell data using cell-state distribution representations (Aim 1) and novel specialized single-cell data representation (Aim 2). Our approach will uncover complex mechanisms of cell differentiation and analyze new modalities like perturb-seq in greater depth. We will demonstrate the utility of our methods with new datasets from collaborations, focusing on hematopoiesis, tissue regeneration and homeostasis in liver, and massive Perturb-seq assays. Aim 1 focuses on developing applications based on Mellon's cell-state distribution representation. This includes tools for differential cell-state abundance, massive relational analysis of Perturb-seq data, local co-regulatory mechanism analysis, and inferring cell-differentiation dynamics using partial differential equations. Aim 2 involves developing computational representation of tissue using cell-state distribution, enabling online learning, sampling strategies, and meaningful gradient estimates. We also aim to use hyperbolic spaces to reduce the computational footprint without sacrificing biologically meaningful signals. By addressing current limitations in data representation and analysis, this project will provide advanced computational tools for single-cell analysis, leading to a deeper understanding of cell differentiation, disease progression, and insights from large perturbation assays. The outcomes will significantly contribute to developmental biology, regenerative medicine, and disease modeling, paving the way for new therapeutic strategies.