A protocol for reconstructing the dynamics of real-world systems from observational data: Application for establishing a digital proxy of a bioreactor (DIYBOT)

Abstract

This protocol sets out a broadly-applicable framework for empirically reconstructing the dynamics of biosystems from observational data, modeling reconstructed dynamics phenomenologically with systems of ordinary differential equations extracted from the data, and detecting and quantifying causal interactions among system components. The protocol draws from well-established procedures in Nonlinear Time Series Analysis (NLTS) that can be run with existing R packages. The protocol was applied in McLamore et al. (2020)1 to develop a phenomenological digital proxy of a bioreactor (DIYBOT) from real-time sensor data. DIYBOT was used to investigate whether bioreactor dynamics self-corrected in response to contamination spikes of increasing degrees; and consequently, the extent to which corrective human-in-the-loop management might be required.

A Connected World: System-Level Support Through Biosensors

“…Digital proxies require real-time sensor data (structured and/or unstructured) that feed a statistically robust in silico model (commonly from the suite of ML tools). Recent work has demonstrated the use of this approach for monitoring flow-through bioreactors, the protocol and code of which are available in an open access form (57,58). In addition, digital proxies are under development for bioprinting (59) and mammalian cell culture ( 60), but to date, sensors have not been integrated for near-real-time modeling, as promised by purported digital twins (54).…”