Comparative Analysis of Binary Similarity Measures in Mass Spectrometry-Based Metabolomics

• Activity Number: 69 - Statistical Methods in Ecology
• Type: Contributed
• Date/Time: Sunday, August 7, 2022 : 4:00 PM to 5:50 PM
• Sponsor: Section on Statistics and the Environment
• Author(s): Seongho Kim* and Ikuko Kato and Xiang Zhang
• Companies: Wayne State University/Karmanos Cancer Institute and Wayne State University/Karmanos Cancer Institute and University of Louisville
• Keywords: Binary similarity; Metabolomics; Compound Identification

Abstract

Compound identification is a key step in mass spectrometry (MS)-based metabolomics. Many methods have been introduced for compound identification based on mass spectral libraries, in silico fragmentation, fragmentation trees, and machine learning. Its most important procedure is to calculate the similarity between binary fingerprints for both tandem MS spectra and experimental or in silico mass spectral libraries. The objective of this study is a comprehensive evaluation of binary similarity measures for compound identification in MS-based metabolomics. Eighteen binary similarity measures, including Jaccard, Dice, Sokal-Sneath, Cosine, Simpson measures, were selected to assess their performance in compound identification using the NIST (National Institute of Standards and Technology) library. Although similarity scores were different, the accuracy of compound identification was the same between the Jaccard, Dice, 3W-Jaccard, Sokal-Sneath, and Kulczynski measures, between the Cosine and Hellinger measures, and between the McConnaughey and Driver-Kroeber measures. Among them, the McConnaughey and Driver-Kroeber measures outperformed others, followed by the Fager-McGowan measure.