Inferring cell property distributions from high-throughput measurement data with cell-to-cell variability

Abstract

High-throughput measurements of cell populations produce broad distributions of observables due to inherent cell-to-cell variability. While these distributions contain valuable biological information, extracting the underlying distributions of cellular properties remains challenging. We introduce the likelihood functional, a statistical framework that infers cell property distributions by combining experimental data with deterministic computational models. The method first evaluates the model on a discrete parameter grid. Optimization of the likelihood functional is then possible without repeated model evaluations and yields best-fit property distributions which enable the direct prediction of observable distributions for comparison with experiments. As a demonstration, we apply the framework to real-time deformability cytometry data and infer cytosol and membrane viscosity distributions of human red blood cells.