There have been significant advances in literature mining, allowing for the extraction of target information from the literature. However, biological literature often includes biological pathway images that are difficult to extract in an easily editable format. To address this challenge, this study aims to develop a machine learning framework called the “Extraction of Biological Pathway Information” (EBPI). The framework automates the search for relevant publications, extracts biological pathway information from images within the literature, including genes, enzymes, and metabolites, and generates the output in a tabular format. For this, this framework determines the direction of biochemical reactions, and detects and classifies texts within biological pathway images. Performance of EBPI was evaluated by comparing the extracted pathway information with manually curated pathway maps. EBPI will be useful for extracting biological pathway information from the literature in a high-throughput manner, and can be used for pathway studies, including metabolic engineering.

中文翻译：

---

用于从文献中的生物途径图像中提取信息的机器学习框架


文献挖掘取得了重大进展，允许从文献中提取目标信息。但是，生物文献通常包含难以以易于编辑的格式提取的生物途径图像。为了应对这一挑战，本研究旨在开发一种称为“生物途径信息提取”（EBPI） 的机器学习框架。该框架可自动搜索相关出版物，从文献中的图像中提取生物通路信息，包括基因、酶和代谢物，并以表格格式生成输出。为此，该框架确定了生化反应的方向，并对生物途径图像中的文本进行检测和分类。通过将提取的通路信息与手动整理的通路图进行比较来评估 EBPI 的性能。EBPI 将有助于以高通量方式从文献中提取生物通路信息，并可用于通路研究，包括代谢工程。