Managing experiment-computational workflows in robotic platforms using Directed Acyclic Graphs

MIONE, FEDERICO M.; SILVA, ALEXIS; LUNA, MARTIN; CRUZ BOURNAZOU, MARIANO N.; MARTINEZ, ERNESTO C.

Kyoto

Simposio; 14th International Symposium on Process Systems Engineering; 2022

Robotic platforms can gather informative data sets to accomplish different modeling or optimization goals for bioprocess development by resorting to on-line redesign of multiple parallel experiments. For reproducible data analysis is key to formally represent and manage experimental-computational workflows in high-throughput experimentation by enforcing FAIR principles. To represent workflows of a robotic platform, directed acyclic graphs (or DAGs) are proposed. Computational implementation of DAGs using open-source software (Apache Airflow) not only helps FAIRizing data and experimental protocols but also obliges making explicit all methods, models, assumptions and hyperparameters used to carry out modeling and optimization tasks. Model-based productivity optimization of a bioprocess based on data from nine fed-batch parallel cultivations is used as an example. Data generated in the parallel experiments are first used to re-estimate online the model parameters and the updated model is used to optimize the feeding profile. Managing experimental-computational workflows as DAGs in the Airflow ecosystem using containers is key to foster the use of FAIR principles in modeling and optimization, and to facilitate access/reuse of costly experimental data.