演題概要

1P-22

PDF

機械学習を応用した医薬品天然物原料のメタボロミクス解析と生産性予測の検討

(アステラス製薬)
^o宮本浩平・ 川瀬直樹・ 古川俊太郎・ 平倉穣

Naturally-derived raw materials are commonly used in fermentation processes to produce pharmaceuticals and serve as nutritious components in the culture media. This study employs metabolomics to analyze natual raw materials, whose quality varies due to factors like origin and processing. Understanding the link between quality and productivity is challenging due to their complex composition. Non-targeted metabolomics was used to analyze them and create a productivity model. Techniques like cold methanol extraction and ultrafiltration were employed to capture metabolites. HILIC/MS identified over 300 components, allowing for batch profiling. Low productivity batches were identified as outliers, indicating a correlation between metabolites and productivity. A productivity model was developed using PLS regression, showing high accuracy. Genetic Algorithms refined the model, maintaining accuracy and highlighting the importance of specific metabolites. Normalizing data across LC/MS measurements is crucial, but reproducibility poses a challenge. Ongoing efforts to address challenges will be also discussed, aiming to advance the development of a reliable prediction method for pharmaceutical production.