Relative correction of experimental results as training data are crucial for optimizing culture media using machine learning

Tsukuba, Japan—Cell culture is a fundamental technology used in several fields, from basic research in life sciences to medical, energy, and material industries. The culture medium (a solution comprising various nutrients) used to grow cells determines the culture results. Thus, selecting a medium appropriate for cell culture is essential. Recently, much focus has been placed on developing cell culture media with less effort and without being bound by conventional empirical rules, for which machine learning is used. Nevertheless, only a few study cases have been investigated this, and further investigation is required. Thus, researchers constructed multiple machine learning models for developing culture media to address this concern. Moreover, they investigated the key aspects of machine learning-assisted media optimization.

Researchers cultured cells derived from human cervical cancer in a medium containing 31 ingredients at various concentrations. The medium compositions and cultured cell concentrations were obtained as training data and subjected to four machine learning models. By applying active learning (repeated machine learning and experimental verification), researchers developed a culture medium that could grow higher cell concentrations than the commercially available medium. Results revealed that the relative value correction of the training data was crucial for effective machine learning. Moreover, additional gene expression analysis was performed to investigate how the optimized culture medium changed the biological pathways in the cell.

This research provides insights into the development of culture media using artificial intelligence and the know-how for its practical application. This application will benefit cell culture technology and contribute to industries and academic research.

###
This work was supported by the JSPS KAKENHI Grant-in-Aid for Challenging Exploratory Research (grant number 21K19815).

_{Original Paper}

Title of original paper:
A data-driven approach for cell culture medium optimization

Journal:
Biochemical Engineering Journal

DOI:
10.1016/j.bej.2024.109591

_{Correspondence}

Associate Professor Bei-Wen Ying
Institute of Life and Environmental Sciences, University of Tsukuba

_{Related Link}

Institute of Life and Environmental Sciences