P06-10

Prediction of medium components for bacteria using deep Learning

Ryuhi SATO *, Takuji YAMADA

Department of Life Science and Technology, Institute of Science Tokyo
( * E-mail: sato.r.bh@m.titech.ac.jp )

Isolation and culture of microorganisms, especially bacteria, is an important process for the application and use of the species. However, most of the known bacteria are either difficult to culture or difficult to isolate and culture. With the recent development of metagenomic analysis technology, the species and functions of bacteria have been inferred by extracting DNA of all bacteria from the environment without isolation and culture. While bacterial genome information can now be obtained, the establishment of culture methods is still an urgent task for research and utilization of unknown bacteria or bacteria with low abundance in samples.
The objective of this study is to identify bacteria-selective culture media from genomic information of bacteria with unknown culture conditions. This study is expected to elucidate the selection and estimation of culture conditions and the relationship between bacterial metabolism and culture medium components.
In this study, we are creating a database that links metabolic pathways based on bacterial culture conditions and functional gene information obtained from genome information. We are also developing a deep learning-based method for predicting culture media components that visualizes the correspondence between bacterial gene information and medium components as inputs. In previous studies, bacteria were treated only in terms of phylogenetic and ecological similarity of species based on 16SrRNA, and therefore, the metabolic pathways of bacteria and the genes involved in them were not mentioned. The proposed method differs from previous studies in that it focuses on the functional genes of bacteria. This allows us to predict the culture medium composition with respect to bacteria that have the same metabolic function but are at different positions in the phylogenetic tree, which cannot be predicted by phylogenetic and ecological similarity of species.