iDeepLC: Chemical Structure Information Yields Improved Retention Time Prediction of Peptides with Unseen...

Area of Science:

• Proteomics
• Analytical Chemistry
• Bioinformatics

Background:

• Deep learning models have advanced proteomics, particularly in predicting peptide retention times for liquid chromatography-mass spectrometry (LC-MS).
• Accurate retention time prediction is crucial for matching experimental data to peptide and protein databases, especially with complex data like data-independent acquisition.
• Existing models struggle to predict retention times for chemically modified peptides, particularly novel or unseen modifications.

Purpose of the Study:

• To develop an enhanced deep learning model (iDeepLC) for more accurate prediction of peptide retention times.
• To improve the prediction of retention times for chemically modified peptides, including those with modifications not seen during model training.
• To leverage chemical structural information for enhanced prediction accuracy and generalization.

Main Methods:

• Developed iDeepLC, an enhanced deep learning model building upon the previous DeepLC model.
• Incorporated chemical structural information (SMILES) of modified residues into the model.
• Evaluated iDeepLC's prediction accuracy and generalization performance against the DeepLC model.

Main Results:

• iDeepLC demonstrates overall more accurate predictions compared to the previous DeepLC model.
• iDeepLC exhibits superior generalization performance in predicting retention times for unseen chemical modifications.
• The model effectively utilizes chemical structure information for improved retention time prediction.

Conclusions:

• iDeepLC represents a significant advancement in predicting peptide retention times, especially for modified peptides.
• The model's ability to generalize to unseen modifications using structural data enhances its utility in proteomics research.
• iDeepLC is available as open-source software, facilitating its adoption and further development in the scientific community.