Fig. 3: Performance of EQ and LCI modules.

a The time cost of EQ and flash entropy on a published dataset (MetabolomicsWorkbench ST001794) against different size theoretical library from MassBank, and all methods were tested on a same personal low-memory computer, using a single thread and CPU (N.D. denotes no data), all tests are repeated 100 times, in the box plot the median as a center line, with the box representing the interquartile range (IQR) between the upper and lower quartiles. Whiskers extend to 1.5 times the IQR, and points beyond this range indicate outliers. The blue boxplot on the left represents the performance of EQ, while the yellow boxplot on the right represents the performance of Flash entropy. b Top-20 recall rates of seven methods tested on published dataset. The dark green circle represents the performance of EQ + LCI using the 1–4 level library, the red pentagram represents the performance of EQ + LCI using the MS-DIAL library, the blue square represents the performance of EQ using the MS-DIAL library, the light blue-purple quadrilateral star represents the performance of Spectral Entropy using the MS-DIAL library, the flesh-colored rhombus represents the performance of Lipidsearch, the dark blue triangle represents the performance of MS-DIAL, and the light green cross represents the performance of LipidMatch. c Detial statistics of recall@Top-20 rates of seven methods classified by lipid subclass. The size of the circle reflects the number of annotations for lipid subclass, and the color shade reflects the recall within the subclass. d Accuracy of LCI and LDA retention time prediction algorithm for different lipid subclasses annotated in published datasets. The red bar chart in the upper part represents the accuracy of LCI, while the blue bar chart in the lower part represents the accuracy of the LDA retention time algorithm. e–h FDR of EQ + LCI, by annotating 8923 lipids in four datasets, including RBL-2H3 cells, mixture dataset, human sera, and zebrafish tissues. Herein, the annotations that not only complied with the ECN rule in tolerance 0.5 min but also contained all feature peaks in high intensity to be correct.