Author contributions:A.C. and B.R. developed the computational methods described here. F.G. and M.C.proposed the initial hypothesis and developed the cell culture, immunostaining, and imaging methods described here. All authors contributed to writing the manuscript. Key Words: retina, self-renewal, stem cell, neural development, cell-fate decision, cell-fate choice, computational biology, algorithmic information theory
ABSTRACTUnderstanding how stem and progenitor cells choose between alternative cell fates is a major challenge in developmental biology. Efforts to tackle this problem have been hampered by the scarcity of markers that can predict cell division outcomes. Here we present a computational method based on algorithmic information theory that can analyze dynamic features of living cells over time. Using this method, we asked whether rat retinal progenitor cells (RPCs) display characteristic phenotypes before undergoing mitosis that could foretell their fate. We were able to predict whether RPCs will undergo a self-renewing or terminal division with 99% accuracy, or whether they will produce two photoreceptors or another combination of offspring with 87% accuracy. Our implementation can segment, track and generate predictions for 40 cells simultaneously on a standard PC at 5 minutes/frame. This method could be used to isolate cell populations with specific developmental potential, thus permitting previously impossible investigations.