Systematic pre-annotation explains the “dark matter” in LC-MS metabolomics

There is no consensus on how to interpret the large number of unknown features in untargeted metabolomics, which are sometimes referred as the “dark matter”. Are these features real compounds or artifacts? Understanding this problem is critical to the annotation and interpretation of metabolomics data and future development of the field. We propose a “detectable khipu” model here, to show that compounds exhibit ion group patterns that depend on their abundance. We apply this model to a systematic analysis of 61 representative public datasets from blood LC-MS metabolomics, the most common data type in biomedical studies. The results indicate that majority of abundant features have identifiable ion patterns, and in-source fragments contribute to less than 10% of features. Each dataset detects 1∼2,000 high confidence compounds, over half of which are unknown. The major knowledge gap in LC-MS metabolomics is therefore not the methods of grouping ions or counting fragments, but the identification of unknown compounds.