Cell growth in Lab-on-a-Chip devices controlled by Artificial Intelligence – LOCAI

In this project, the consortium members will develop an automated system for cell culture optimization based on the combination of machine learning with Lab-on-a-Chip (LOC) technology. The benefits of microfluidics, coupled with the ability of machine learning to process the large amounts of data and variables of microfluidic systems, will lead to a unique, easy-to-use tool that will advance automation and precision levels for cell cultures.

This new system will involve real-time monitoring of the state of a cell culture through the analysis of images by a machine learning algorithm. These parameters will then be provided to another agent, which will control the pumps that inject medium and reagents into the culture. Thereby, a closed-loop system completely controlled by machine learning is obtained, which will allow the automatic analysis of thousands of variables in a single test.

To achieve this goal, the system will be based on the use of two main areas of machine learning. With supervised machine learning, annotated images of the cell cultures will be used to train an artificial neural network to automatically detect and distinguish individual cells from the image background. The individual detections allow to derive several accurate measurements such as cell count, cell area, morphology and other metrics, required for an accurate estimate of the cell culture state.

’Reinforcement learning’ is an area of machine learning which aims to learn optimal behavior in an environment with unknown dynamics. In the presented system, an algorithm will combine the state information extracted from the image analysis and the possibility to modulate the medium composition in order to optimize the state of the cell culture with respect to a user-defined metric (e.g., total cell area). The algorithm monitors the culture’s state while continuously re-adjusting the medium composition. Thereby, it learns about relations between cell culture state and media composition and will ultimately be able to derive a time and state-dependent behavior that optimizes the user-defined metric, resulting in an optimized cell culture growth as well as the corresponding protocol with which to achieve the former.
Although this tool will be very valuable for all types of cultures, as particular applications and proof of concept for suspension as well as adherent cells, this system will be tested for the maintenance and growth of two central players of the immune response in the body and the central nervous system (CNS). Specifically, we plan to culture autologous T-cells (suspension cells) where the algorithm will optimize their production by defining the cytokine cocktail composition and schedule of administration; and culture microglial cells (adherent cells), the tissue-resident macrophages of the CNS, to assess neuro-inflammatory processes and therapies in neurodegenerative pathologies.