Mechanistic ordinary differential equation models of gene regulatory networks are a valuable tool for understanding biological processes that occur inside a cell, and they allow for the formulation of novel hypotheses on the mechanisms underlying these processes. Although data-driven methods for inferring these mechanistic models are becoming more prevalent, it is often unclear how recent advances in machine learning can be used effectively without jeopardi zing the interpretability of the resulting models. In this work, we present a framework to leverage neural networks for the identification of data-driven models for time-dependent intracellular processes, such as cell differentiation. In particular, we use a graph autoencoder model to suggest novel connections in a gene regulatory network. We show how the improvement of the graph suggested using this neural network leads to the generation of hypotheses on the dynamics of the resulting identified dynamical system.
Klomp et al. (Fri,) studied this question.