Breaking Down Structural Diversity for Comprehensive Prediction of Ion-Neutral Collision Cross Sections

Area of Science:

• Analytical Chemistry
• Computational Chemistry
• Biochemistry

Background:

• Accurate identification of unknown compounds is essential for untargeted analyses like metabolomics and drug metabolite identification.
• Ion mobility-mass spectrometry (IM-MS) separates ions based on collision cross section (CCS), a size- and shape-dependent property valuable for compound identification.
• Current limitations in identifying unknowns stem from the scarcity of high-quality reference CCS values and the need for broadly applicable CCS prediction models.

Purpose of the Study:

• To develop a novel machine learning (ML) approach for accurate CCS prediction across diverse chemical structures.
• To identify key structural characteristics influencing CCS variance and evaluate ML algorithms for CCS prediction.
• To create a publicly accessible resource for querying CCS data and utilizing predictive models.

Main Methods:

• Utilized a comprehensive database of CCS values covering a wide chemical space.
• Employed molecular quantum numbers (MQNs) to represent chemical structural diversity.
• Applied unsupervised clustering based on MQNs to group similar chemical structures.
• Trained and evaluated various ML algorithms for CCS prediction within identified clusters.

Main Results:

• Identified MQNs as significant contributors to CCS variance.
• Demonstrated that ML models trained on clustered data significantly outperform a single model trained on all data.
• Achieved high accuracy in CCS prediction for diverse chemical structures using the developed ML approach.
• Developed CCSbase.net, a web interface for accessing the CCS database and predictive model.

Conclusions:

• The ML approach utilizing MQN-based clustering provides accurate and robust CCS predictions.
• This method enhances the identification of unknown compounds in large-scale untargeted analyses.
• CCSbase.net offers a valuable tool for researchers in metabolomics, drug discovery, and other fields requiring compound identification.