Hiroki Shioda (D2), Department of Chemical System Engineering, received Best Poster Award -  Honorable Mention, The 24th Asia Pacific Bioinformatics Conference

On 29th October 2026, Hiroki Shioda (D2), Sakai Y and Nishikawa M Lab, Department of Chemical System Engineering, received Best Poster Award – Honorable Mention, The 24th Asia Pacific Bioinformatics Conference.

Best Poster Award – Honorable Mention, The 24th Asia Pacific Bioinformatics Conference
This award was established to recognize outstanding poster presentations at the Asia - Pacific Bioinformatics Conference  (APBC 2026). Posters were evaluated based on their research content, originality, and academic significance, and this work was selected for the honorable mentions of the Best Poster Award.

About awarded research
“Construction of an endothelial cell differentiation protocol database via sequential guided extraction using large language models”
To  analyze and optimize cell culture protocols, cross paper meta-analysis could be highly beneficial. However, collecting data about cell culture protocols from papers is a complex process. It requires interpreting information scattered throughout texts, while considering the overall context.
Large language models (LLMs) can extract information from documents as structured data. However, their performance declines with increased input size. In this study, we developed a novel workflow that enables LLMs to comprehend long texts by dividing them into segments while also retaining paper-specific context. We applied this approach to papers on endothelial cell differentiation to create a database of differentiation methods and analyze their relationship with resulting cellular phenotypes.
In the constructed workflow, first, texts surrounding keywords related to cell culture methods were extracted and arranged in their order of appearance in the original text. To exclude irrelevant text segments such as information of cited papers, relevance of text segments were judged by majority voting across five independent LLM evaluations. The relevant segment was then provided to the LLM to obtain dictionary-format output about methods and results of differentiation, such as cell type, gene expression, culture media, and additives. Data obtained from text segments were also used as input to provide information about paper-specific context and terminology in the extraction of subsequent segments.
We applied this method to 362 publications on in vitro endothelial cell differentiation identified using Web of Science, and constructed a database containing their structured protocol data.  Classification using the SoftMax function revealed a trend in endothelial cell phenotypes. Retinoic acid was found to be more effective in acquiring a brain microvascular phenotype than an arterial or progenitor phenotype, consistent with previous reports. Furthermore, we implemented a web-based database (https://ituke.github.io/cellcultureatlas/) that enables users to search protocols based on cell types, gene expression, and other characteristics.
In summary, we developed a novel data extraction method and used it to perform a detailed meta-analysis of endothelial differentiation methods. We improved extraction accuracy by dividing papers into appropriately sized text segments and processing them sequentially to preserve context. The relationships between culture conditions and phenotypes extracted from literature were consistent with previous reports. This approach could be applied to literature mining in various fields where key information is scattered across papers, and could also support future data-driven protocol design.

Your impression & future plan
I would like to express my sincere gratitude to the co-authors of this project, Mr. Shusuke Ito and Mr. Gaku Shimodaira (Department of Chemical System Engineering) who contributed equal amounts to the research. I would also like to express my sincere gratitude to Professor Yasuyuki Sakai, Koki Muraoka and Masaki Nishikawa for their comments and feedback that greatly improved this research. 
Cell culture method optimization still relies heavily on manual trial and error, compared to the analysis of results which utilizes various data driven methods such as omics analysis. Encouraged by this award, I will continue to continue on this research that aims to bring in a more data driven approach to cell culture method analysis.