in The Ohio Union

Approaches for Dealing with Data Disparity and Complex Dynamics

Monday, July 17 at 6:00pm

SMB2023 SMB2023 Follow Monday during the "PS01" time block.
Room assignment: in The Ohio Union.
Share this

Ari Barnett (Roldan)

Moffitt Cancer Center
"Approaches for Dealing with Data Disparity and Complex Dynamics"
Data disparity remains a persistent challenge for the broader translational science community. At present, models working with observational data frequently encounter difficulties stemming from inconsistent measurement frequencies and insufficiently diverse patient populations. Approaching this as a compounded problem we seek to develop a novel framework that utilizes the concept of Time series Generative Adversarial Networks (TGAN) originally proposed by Yoon [1]. While generative frameworks have been introduced, none can fully provide a sound solution for the temporal dynamics involved with time series observations. TGAN specifically aims to address temporal dynamics by utilizing a jointly optimized embedding space. Here we propose utilizing TGAN to generate both synthetic patients and semi-synthetic time series. Previously TGAN has been shown to outperform similar approaches, both qualitatively (tSNE) and quantitatively (discriminative and predictive scoring) on a variety of real-world datasets. For this research we aim to provide a conceptual methodology for aiding in the discovery of underlying mechanistic models via the integration of SINDy [2].By utilizing synthetic data that capture underlying dynamics we hypothesize that we can train models while holding out all real observation data for testing. Similarly with semi-synthetic time series we anticipate a better overall capture of disease dynamics. References [1] J. Yoon, D. Jarrett, and M. van der Schaar, “Time-series Generative Adversarial Networks,” in Advances in Neural Information Processing Systems, Curran Associates, Inc., 2019. Accessed: Feb. 14, 2023. [Online]. Available: [2] S. L. Brunton, J. L. Proctor, and J. N. Kutz, “Discovering governing equations from data by sparse identification of nonlinear dynamical systems,” Proc. Natl. Acad. Sci. U.S.A., vol. 113, no. 15, pp. 3932–3937, Apr. 2016, doi: 10.1073/pnas.1517384113.
Additional authors: Renee Brady-Nicholls (Moffitt Cancer Center)

#SMB2023 Follow
Annual Meeting for the Society for Mathematical Biology, 2023.