ReTimeML: a retention time predictor that supports the LC-MS/MS analysis of sphingolipids

Area of Science:

• Lipidomics
• Mass Spectrometry
• Bioinformatics

Background:

• Analyzing ceramide (Cer) and sphingomyelin (SM) lipid species via LC-MS/MS is challenging due to ambiguous mass and fragmentation patterns.
• Existing methods struggle to differentiate between multiple molecular arrangements with similar mass-to-charge ratios.

Purpose of the Study:

• To develop a computational tool, ReTimeML, for automating the prediction of retention times (RTs) for Cer and SM lipid species.
• To enhance the accuracy and efficiency of lipidomic analyses by resolving ambiguities in LC-MS/MS data.

Main Methods:

• ReTimeML utilizes machine learning to build a regression library correlating mass-to-charge (m/z) with RTs from known standards.
• The software extrapolates RTs for unknown lipid species based on established RTs from internal standards, calibrators, or quality controls.
• The model does not require retraining for different LC-MS/MS pipelines, ensuring broad applicability.

Main Results:

• ReTimeML demonstrated high accuracy in RT estimations for diverse Cer and SM structures across various biological samples and LC-MS/MS setups, with a mean variance of 0.23–2.43% compared to manual annotations.
• The tool successfully disambiguated SM identities from isobaric interferences in serum and cerebrospinal fluid samples.
• Novel, non-canonical sphingomyelins with polyunsaturated structures, exhibiting enhanced stability against catabolic clearance, were identified.

Conclusions:

• ReTimeML offers a robust and adaptable solution for automating lipid retention time prediction in complex LC-MS/MS analyses.
• The software improves the reliability of ceramide and sphingomyelin identification, particularly in challenging datasets.
• This advancement facilitates the discovery of previously uncharacterized lipid species and their biological implications.