可解释性
计算机科学
形态发生剂
形式主义(音乐)
连锁
过程(计算)
拓扑(电路)
动力系统理论
基因调控网络
插值(计算机图形学)
理论计算机科学
简单(哲学)
人工智能
变形
生物
程式化事实
健身景观
瓶颈
坐标系
人工神经网络
数学
复杂系统
作者
Dillon Cislo,M. Joaquina Delás,James Briscoe,Eric D. Siggia
标识
DOI:10.1073/pnas.2521762122
摘要
The development of a zygote into a functional organism requires that this single progenitor cell gives rise to numerous distinct cell types. Attempts to exhaustively tabulate the interactions within developmental signaling networks that coordinate these hierarchical cell fate transitions are difficult to interpret or fit to data. An alternative approach models the cellular decision-making process as a flow in an abstract landscape whose signal-dependent topography defines the possible developmental outcomes and the transitions between them. Prior applications of this formalism have built landscapes in low-dimensional spaces without explicit maps to gene expression. Here, we present a computational geometry framework for fitting dynamical landscapes directly to high-dimensional single-cell data. Our method models the time evolution of probability distributions in gene expression space, enabling landscape construction with minimal free parameters and precise characterization of dynamical features, including fixed points, unstable manifolds, and basins of attraction. We demonstrate the applicability of this framework to multicolor flow-cytometry and RNA-seq data. Applied to a stem cell system that models ventral neural tube patterning, we recover a family of morphogen-dependent landscapes whose valleys align with canonical neural progenitor types. Remarkably, simple linear interpolation between landscapes captures signaling dependence, and chaining landscapes together reveals irreversible behavior following transient morphogen exposure. Our method combines the interpretability of landscape models with a direct connection to data, providing a general framework for understanding and controlling developmental dynamics.
科研通智能强力驱动
Strongly Powered by AbleSci AI