Sitetack: a deep learning model that improves PTM prediction by using known PTMs.

Area of Science:

• Proteomics
• Bioinformatics
• Computational Biology

Background:

• Post-translational modifications (PTMs) are crucial for proteome diversity and have therapeutic implications.
• Deep learning models are increasingly used for predicting PTM sites.
• Current prediction models face limitations due to dataset constraints and analysis methods.

Purpose of the Study:

• To evaluate the impact of known PTM sites on the accuracy of sequence-based deep learning algorithms for PTM prediction.
• To investigate whether knowledge of one PTM's location can improve the prediction of other PTMs.

Main Methods:

• Developed sequence-based deep learning models using convolutional neural networks.
• Encoded known PTM locations as distinct amino acids within sequences.
• Utilized word embedding for sequence representation.
• Compared model performance with and without labeling known PTM sites.

Main Results:

• Models achieved performance comparable to existing methods without labeled PTM data.
• Labeling known PTM sites led to significant improvements over existing models.
• Knowledge of PTM locations enhanced the predictability of other PTMs.

Conclusions:

• The inclusion of known PTM locations substantially boosts the performance of deep learning models for PTM prediction.
• PTMs play a critical role in the occurrence of subsequent PTMs.
• This approach is expected to enhance the performance of various proteomic machine learning algorithms.