Machine Learning-Based Retention Time Prediction Tool for Routine LC-MS Data Analysis

Accurate retention time (RT) prediction
models can significantly improve liquid chromatography–mass
spectrometry (LC-MS) data analysis widely used in chemical synthesis.
As hundreds of thousands of syntheses are performed annually at Enamine,
a large amount of experimental data has been generated internally.
In this paper, we present the development of an RT prediction model based on the GATv2Conv + DL graph neural
network (NN) architecture, trained on the internal data and further
evaluated using the METLIN SMRT data set. The final model achieved
a mean absolute error (MAE) of 2.48 s for the 120 s LC-MS method.
We also conducted a detailed analysis of RT prediction errors and determined that the interval between RT – 7.12 s and RT + 9.58 s contained over 95% of the data. The developed model
has been successfully integrated into the existing in-house LC-MS
analysis toolkit, enhancing its predictive and analytical capabilities.
Additionally, we have published a curated subset of 20,000 data points
from our internal data set to support community benchmarking and further
research.