Improved Peptide Retention Time Prediction in Liquid Chromatography through Deep Learning.

Area of Science:

• Proteomics
• Computational Biology
• Analytical Chemistry

Background:

• Peptide retention time (RT) prediction accuracy is crucial for proteomics but currently insufficient.
• Existing models lack the precision needed for widespread adoption in proteomic workflows.

Purpose of the Study:

• To develop a deep learning model for substantially improving peptide RT prediction accuracy.
• To introduce DeepRT, a novel tool leveraging capsule networks for enhanced RT prediction.

Main Methods:

• Designed DeepRT using a capsule network architecture.
• Evaluated DeepRT performance on public datasets from reverse-phase liquid chromatography (LC).
• Assessed DeepRT's adaptability to strong cation exchange (SCX) and hydrophilic interaction liquid chromatography (HILIC).

Main Results:

• DeepRT achieved high prediction accuracy with R² values around 0.994 for reverse-phase LC.
• Demonstrated strong performance on SCX (R² ≈ 0.996) and HILIC (R² ≈ 0.993) separations.
• DeepRT(+) model, utilizing transfer learning, significantly improved predictions for small datasets.

Conclusions:

• DeepRT offers a flexible and efficient solution for accurate peptide RT prediction.
• The model learns fundamental amino acid retention properties, aligning with established coefficients.
• DeepRT represents a significant advancement for proteomics data analysis and implementation.