- Timetable
- Download all abstracts
- Plenary Lectures
- MSSJ Special Program
- Award Lecture
- Symposium Sessions（Day1, Day2）
- Fundamental Sessions（Day1, Day3）
- Young Researchers' Sessions (Int'l)（Day1, Day3）
- Young Researchers' Sessions（Day1, Day3）
- Poster Presentations（Day1, Day2, Day3）
- Evening Workshop
- Corporate Program
Symposium Sessions
Day 2, June 11（Tue.） 9:55-10:15 Room B (Convention Hall 200)
- 2B-S-0955
Tandem Mass Spectra Machine Learning for Lipid Subclass Prediction
(1Tokyo Univ. Agr. Tech., 2RIKEN IMS, 3Keio Univ., 4Yokohama City Univ.)
oNami Sakamoto1, Takaki Oka1, Yuki Matsuzawa1, Kozo Nishida1, Aya Hori2, Makoto Arita2,3,4, Hiroshi Tsugawa1,2,4
Untargeted lipidomics using collision-induced dissociation based tandem mass spectrometry (CID-MS/MS) is an essential technique in biology and clinical application. However, the annotation confidence is still guaranteed by the manual curation of analytical chemists although various software tools have been developed for the automatic spectral processing based on the rule-based fragment annotations. In this study, we developed MS2lipid, a machine learning model, to predict lipid subclasses learning 82,455 curated spectra obtained from 82 datasets. MS2lipid achieved over 94.6% accuracy for 97 lipid subclasses. Moreover, our program outperformed the accuracy of CANOPAS ontology prediction in which ours provided 35.9% higher value of F1 score on average. Furthermore, the MS2lipid program offers over 87% accuracy on average for spectra acquired by different curators and different MS techniques. Furthermore, the function of MS2lipid was showcased by the annotation of novel esterified bile acids in combination with molecular spectrum networking. Consequently, our machine learning model provides an independent criterion for lipid subclass classification in addition to an environment for annotating lipid metabolites that have been previously unknowns.