Illuminating the Robust and Precise Developmental Patterning with Deep-Learning Embryos


Feng Liu, School of Physics, Center for Quantitative Biology, Peking University


2018.05.15 14:00-15:00


601, Pao Yue-Kong Library


“More is different”, said by P.W. Anderson in 1972. Now the frontier of physical biology has been extended to studying the collective cell behavior in multicellular organisms from a quantitative, holistic perspective. One of the most intriguing questions in this field is how macroscopic patterns in developing embryos emerge from the dynamic interaction between individual cells. In particular, how does the multicellular system achieve the high precision and reproducibility in patterning despite the intrinsic noise as well as extrinsic environmental fluctuations? Combining quantitative imaging and modeling, we have investigated the mechanism of pattern formation in fruit fly embryos. My talk will focus on modeling the early developmental patterning. The spatiotemporal developmental patterns have been extensively measured, however, the underlying mechanism remains elusive because of the daunting task to build a quantitative gene network model to account for the rich precise measurement results. We have developed a deep neuron network (DNN) model and trained it with the dynamical developmental expression profiles of wild type flies. For the first time, the prediction of this model is consistent with the measured developmental expression profiles in maternal or gap gene mutants, the dynamical correction of the gap gene profiles under morphogen Bicoid dosage perturbation, and the scaling of the gap gene profiles when the embryo size varies. Furthermore, the gene network extracted from DNN recovers the widely accepted strong regulation and suggests some potentially important context-dependent regulation. This reverse engineering approach in coupled with a forward engineering modeling at the enhancer level could help to direct further measurements, and reveal how the robust and precise pattern formation in the developing fruit fly embryos is achieved through dynamic integration of inputs from multiple maternal morphogen gradients and collective synergy of the segmentation gene network. Our work demonstrates the potential utility of applying physics approaches based on precise measurements and quantitative modeling to uncover the design principle of biological network orchestrating the emerged collective multicellular behavior.


Dr. Feng Liu is an Assistant Professor in the School of Physics and the Center for Quantitative Biology at Peking University. He received his Ph.D. in Biophysics and Computational Biology from University of Illinois at Urbana and Champaign, MS in Nuclear Engineering from Peking University, and BA in Physics from Shandong University. His expertise lies in quantitative systems biology, biophotonics, and computational biology. His Ph.D. work was focused on protein folding, protein aggregation and molecular dynamics simulations. The proteins designed by him could fold with extremely fast speed (reaching the folding limit) and were served as benchmark proteins for molecular dynamics simulations. During his postdoc work at Princeton University, he studied the pattern formation during Dorsophila embryogenesis using quantitative two-photon microscopy, genetics and mathematical modeling. His work on quantitatively testing threshold dependent model in cell fate determination was published in PNAS and highlighted by Nature Reviews Genetics. His current work focuses on developing high-speed 3D imaging tools and computational models to investigate the gene network regulating in embryogenesis and cancer metastasis. Dr. Feng Liu has published a series of papers in prestigious journals such as Nature Communications, PNAS, Advanced Functional Materials, Current Biology and etc.