Fig. 4: Performance of WMYn.

a Sea stack plots of entropy similarity between WMYn predictions and GroundTruth (1st injection of lipid reference), 2nd injection of lipid references, and 3rd injection of lipid reference, statistically based on an interval length of 0.05, with the x-axis denoting the number of intervals within each interval. GroundTruth, Reference 2, and Reference 3 are lipid reference standards (9 in positive ion mode and 6 in negative ion mode) continuously collected three times on Orbitrap Exploris 240 MS system. b Retention time deviation of LipidIN framework annotations and GroundTruth, Reference 2, and Reference 3, confirming the accuracy of the LipidIN framework. We used 15 Lipid reference data from three injections to perform a bilateral t-test, and error bars are drawn using the mean and standard deviation (SD) to represent variability in the data. c. Platform transfer performance of WMYn from Orbitrap Exploris 240 MS system to Agilent qTOF MS system. d–e Entropy similarity comparison of WMYn predictions with other fitting methods, including mean method, linear fitting, polynomial fitting, and exponential fitting in low-resolution MS data with two decimals and high-resolution MS data with four decimals. f Theoretical spectra from MS-DIAL public database and reverse lipidomics spectra were tested separately in Entropy Search environment. The recall@Top1 at different thresholds was counted separately, and colors were used to distinguish the two libraries (red: with 5th-level library, blue: without 5th-level library). g The recall@Top-1 of different lipid classes under two libraries was counted separately at a modest cutoff threshold at 0.75 similarity score.