Computing That Serves

Data-Driven Methods for Discovering Novel Biological Computing Architectures


Thursday, March 15, 2018 - 11:00am


Enoch Yeung


David Wingate

Colloquium presented by Enoch Yeung
Thursday, March 15, 2018 at 11:00 A.M.
Location: 1170 TMCB

Biological computing is ever present in our world, from the complex nanoscale interactions of microbial consortia to macro-scale social dynamics of ecosystems. Populations of millions of cells within a palm-sized test-tube implement stochastic and massively parallel computing, reprogramming their own DNA to manifest diverse computing architectures over time. Amazingly, individual prokaryotic cells achieve power efficiency and robustness at a scale that has yet to be matched in silicon based systems.   To discover, engineer, and imbue such properties into biomimetic and biological computing systems is the challenge of our time.  In this talk I will overview recent computational advances in data science and system identification that accelerate design of biological and biomimetic computing.  I will highlight two advances in particular: 1) the discovery of a novel biophysical mechanism for implementing feedback control in synthetic gene networks and 2) the integration of deep learning and data-driven methods to generate a computational framework for predicting the operational envelopes of synthetic biocircuits.   In the former, I will discuss how overlooked design properties of synthetic gene networks can be used to enforce local and fast feedback control, resulting in improved persistence (over 72 cell divisions) of memory in the genetic toggle switch.   In the latter, I address the challenge of predicting biological system behavior over a range of conditions, given finite and noisy experimental samples.    Specifically, we examine the problem of discovering models for moment dynamics of a biological network, as a function of small molecule control signals and temperature-based perturbations.  We introduce deep dynamic mode decomposition and show that the input-Koopman models learned from flow cytometry data predict biocircuit functionality with over 90% accuracy over 1,000 experimental conditions.  I conclude this talk underscoring open problems in data-driven learning and design for complex systems, where canonical models are lacking or underlying complexity prohibits traditional analytical approaches.   



Dr. Enoch Yeung is a senior research scientist at the Pacific Northwest National Laboratory (PNNL). He has a Ph.D. in Control and Dynamical Systems from the California Institute of Technology and a B.S. in Mathematics, magna cum laude, from Brigham Young University. He is currently the principal investigator for projects on distributed biological computing (HPDA), design of data-driven control architectures (PNNL Control of Complex Systems Initiative) and data-driven design and discovery for synthetic biological systems (DARPA SD2). He has lead major research tasks on several multi-institutional collaborative synthetic biological research programs including the DARPA Living Foundries program, the NSF Molecular Programming Project, and the AFOSR Biological Research Initiative. He as served on several advisory panels for DARPA, NIST, and the NDU. He is the recipient of Kanel Foundation Fellowship, the National Science Foundation Graduate Fellowship, a National Defense Science and Engineering Fellowship, the PNNL Project Team of the Year Award, and the PNNL Outstanding Performance Award.