DeepDigest

DeepDigest predicts proteotypic cleavage sites for proteins digested by trypsin, ArgC, chymotrypsin, GluC, LysC, AspN, LysN, and LysargiNase to model protease digestion behavior and improve peptide identification in shotgun proteomics.

Key Features:

• Supported proteases: Handles trypsin, ArgC, chymotrypsin, GluC, LysC, AspN, LysN, and LysargiNase.
• Model architecture: Uses a sequence-based deep learning approach integrating convolutional neural networks (CNNs) and long short-term memory networks (LSTMs).
• Cleavage probability prediction: Predicts cleavage probabilities at potential cleavage sites and models missed cleavage events.
• Protease-specific characterization: Provides insights into unique digestion behaviors of different proteases.
• Performance evaluation: Outperformed logistic regression, random forest, and support vector machine baselines with 10-fold cross-validation AUCs of 0.956–0.982 on eight training datasets and AUCs of 0.849–0.978 on eleven independent test datasets.
• Transfer learning: Employs transfer learning to enhance prediction accuracy across datasets.
• Peptide digestibility: Predicts peptide digestibility and can distinguish correct from incorrect peptide identifications.

Scientific Applications:

• Experimental design: Inform selection of proteases and digestion strategies in shotgun proteomics experiments by modeling protease-specific cleavage behavior.
• Peptide identification: Improve peptide identification accuracy by providing predicted proteotypic cleavage sites and peptide digestibility scores.
• Protease behavior analysis: Enable comparative analysis of digestion patterns across multiple proteases.

Methodology:

Sequence-based deep learning integrating CNNs and LSTMs; model comparison to logistic regression, random forest, and support vector machine; evaluation via 10-fold cross-validation on eight training datasets and testing on eleven independent datasets; transfer learning applied to boost accuracy.