Skip to main content

Table 5 Percentage of results where the true molecule was ranked among the top three candidates using the meta-learning method

From: Probabilistic metabolite annotation using retention time prediction and meta-learned projections

CM No. training points 10 20 30 40 50
FEM long \(71.07\pm 5.11\) \(72.75\pm 4.40\) \(74.42\pm 5.29\) \(75.27\pm 2.25\) \(76.10\pm 2.92\)
\((59.57 \pm 6.73)\)      
FEM orb. \(64.96\pm 10.96\) \(67.13\pm 10.04\) \(65.25\pm 8.41\) \(63.06\pm 9.07\) \(67.04\pm 9.40\)
\((53.57\pm 7.66)\)      
LIFE old \(62.99\pm 5.21\) \(61.38\pm 5.13\) \(60.50\pm 6.68\) \(58.57\pm 4.83\) \(59.86\pm 3.36\)
\((51.44\pm 7.26)\)      
RIKEN \(72.89\pm 8.59\) \(73.21\pm 4.37\) \(74.44\pm 4.17\) \(77.61\pm 4.04\) \(75.73\pm 3.81\)
\((53.62\pm 5.77)\)      
  1. Reported percentages in [21] were 66.7%, 67.9%, 69.7% and 71.9% for FEM long, FEM orbitrap plasma (Fem orb. in the table), LIFE old and RIKEN, respectively. Values between parentheses under the name of the Chromatographic Method (CM) represent the baseline performance when only mass information is used