Network Science Institute, Northeastern University, Boston, MA, USA.
Division of Network Medicine, Department of Medicine, Harvard Medical School, Boston, MA, USA.
Sci Rep. 2020 Oct 1;10(1):16191. doi: 10.1038/s41598-020-73105-0.
Thanks to the many chemical and nutritional components it carries, diet critically affects human health. However, the currently available comprehensive databases on food composition cover only a tiny fraction of the total number of chemicals present in our food, focusing on the nutritional components essential for our health. Indeed, thousands of other molecules, many of which have well documented health implications, remain untracked. To explore the body of knowledge available on food composition, we built FoodMine, an algorithm that uses natural language processing to identify papers from PubMed that potentially report on the chemical composition of garlic and cocoa. After extracting from each paper information on the reported quantities of chemicals, we find that the scientific literature carries extensive information on the detailed chemical components of food that is currently not integrated in databases. Finally, we use unsupervised machine learning to create chemical embeddings, finding that the chemicals identified by FoodMine tend to have direct health relevance, reflecting the scientific community's focus on health-related chemicals in our food.
由于其携带的许多化学和营养成分，饮食对人类健康至关重要。然而，目前可用的综合性食物成分数据库仅涵盖了我们食物中存在的化学物质的一小部分，重点是对我们健康至关重要的营养成分。事实上，还有数千种其他分子，其中许多分子对健康有明确的影响，仍然没有被追踪到。为了探索有关食物成分的现有知识体系，我们构建了 FoodMine，这是一种使用自然语言处理技术从 PubMed 中识别可能报告大蒜和可可化学组成的论文的算法。从每篇论文中提取有关报告化学物质数量的信息后，我们发现科学文献中载有大量有关食物详细化学成分的信息，而这些信息目前并未整合到数据库中。最后，我们使用无监督机器学习来创建化学嵌入，发现 FoodMine 识别的化学物质往往与直接的健康相关性有关，反映了科学界对我们食物中与健康相关的化学物质的关注。
