Inferring single-cell protein levels and cell cycle behavior in heterogeneous cell populations

Abstract

Individual cells in a genetically identical population can show highly variable behavior. Single-cell measurements allow us to study this variability, but the available measurement techniques have limitations: live-cell microscopy is typically restricted to one or a few molecular markers, while techniques that simultaneously measure large numbers of molecular markers are destructive and cannot be used to follow cells over time. To help overcome these limitations, we present here scMeMo (single cell Mechanistic Modeler): a mechanistic modeling framework that can leverage diverse sets of measurements in order to infer unobserved variables in heterogeneous single cells. We used this framework to construct a model describing cell cycle progression in human cells, and show that it can predict the levels of several proteins in individual cells, based on live-cell microscopy measurements of only one marker and information learned from other experiments. The framework incorporates an uncertainty calibration step that makes the posterior distributions robust against partial model misspecification. Our modeling framework can be used to integrate information from separate experiments with diverse readouts, and to infer single cell variables that may be difficult to measure directly.