Penalized inference of the hematopoietic cell differentiation network via high-dimensional clonal tracking

Background: During their lifespan, stem- or progenitor cells have the ability to differentiate into more committed cell lineages. Understanding this process can be key in treating certain diseases. However, up until now only limited information about the cell differentiation process is known. Aim: The goal of this paper is to present a statistical framework able to describe the cell differentiation process at the single clone level and to provide a corresponding inferential procedure for parameters estimation and structure reconstruction of the differentiation network. Approach: We propose a multidimensional, continuous-time Markov model with density-dependent transition probabilities linear in sub-population sizes and rates. The inferential procedure is based on an iterative calculation of approximated solutions for two systems of ordinary differential equations, describing process moments evolution over time, that are analytically derived from the process’ master equation. Network sparsity is induced by adding a SCAD-based penalization term in the generalized least squares objective function. Results: The methods proposed here have been tested by means of a simulation study and then applied to a data set derived from a gene therapy clinical trial, in order to investigate hematopoiesis in humans, in-vivo. The hematopoietic structure estimated contradicts the classical dichotomy theory of cell differentiation and supports a novel myeloid-based model recently proposed in the literature.