Development of a Predictive Multiple Reaction Monitoring (MRM) Model for High-Throughput ADME Analyses Using...

Area of Science:

• Analytical Chemistry
• Computational Chemistry
• Pharmacology

Background:

• Multiple Reaction Monitoring (MRM) is a key technique in drug discovery for quantifying compounds.
• Experimental optimization of MRM conditions, including product ion selection, is time-consuming and resource-intensive.
• Predicting product ions computationally can significantly accelerate the drug development pipeline.

Purpose of the Study:

• To develop a Learning-to-Rank (LTR) model for predicting product ions directly from compound structures.
• To eliminate the need for experimental MRM optimization in drug discovery workflows.
• To assess the model's performance and its impact on high-throughput biological assays.

Main Methods:

• Utilized a dataset of 5757 experimentally determined MRM conditions.
• Generated theoretical fragments and mass-to-charge ratios using MassChemSite software.
• Trained and validated LTR models using metrics like NDCG@k and Coverage@k.

Main Results:

• The LTR model achieved high performance on a validation set (NDCG@5 = 0.732, Coverage@5 = 0.841).
• Predictions from the model were applied to high-throughput Caco-2 permeability and metabolic stability assays.
• The model's predictions resulted in 97% biologically equivalent outcomes compared to experimental MRM conditions.

Conclusions:

• The developed LTR model effectively predicts product ions for MRM analysis, reducing experimental workload.
• This computational approach offers a significant advancement for accelerating drug discovery and development.
• The model's successful application in biological assays demonstrates its practical utility in quantitative mass spectrometry.