Predicting Ion Mobility Collision Cross-Sections Using a Deep Neural Network: DeepCCS.

Area of Science:

• Analytical Chemistry
• Computational Chemistry
• Biochemistry

Background:

• Untargeted metabolomics using mass spectrometry is vital for discovering biologically important small molecules.
• Accurate small molecule identification is challenging due to fragmentation data limitations and reliance on databases or NMR.
• Gas-phase collision cross section (CCS) from ion mobility spectrometry (IMS) improves metabolite identification but lacks sufficient empirical data.

Purpose of the Study:

• To develop a predictive model for CCS values using deep learning to overcome data limitations in IMS.
• To create a computational tool that requires only SMILES notation and ion type for CCS prediction.
• To enhance the accuracy and efficiency of small molecule identification in metabolomics.

Main Methods:

• A deep learning algorithm was developed and trained to predict CCS values.
• The model utilized compound SMILES notation and ion type as input.
• Training and testing involved over 2400 molecules from five different laboratories and instruments.

Main Results:

• The deep learning model achieved a coefficient of determination (R²) of 0.97.
• The median relative error for CCS predictions was 2.7% across a diverse range of molecules.
• The developed method demonstrated superior performance compared to existing CCS prediction algorithms.

Conclusions:

• The deep learning model provides accurate and efficient CCS predictions for small molecules.
• This approach significantly reduces the challenge of small molecule identification in untargeted metabolomics.
• The model requires minimal computational resources, making it broadly applicable.