Protein feature engineering framework for AMPylation site prediction.

Area of Science:

• Biochemistry
• Bioinformatics
• Computational Biology

Background:

• AMPylation is a significant post-translational modification involving the addition of adenosine monophosphate (AMP) to tyrosine and threonine residues.
• While its prevalence and impact are increasingly understood, experimental identification of AMPylation sites is difficult.
• Computational prediction offers a faster alternative, but model performance relies heavily on feature representation.

Purpose of the Study:

• To develop a novel feature extraction pipeline for improved computational prediction of AMPylation sites.
• To evaluate the effectiveness of these novel features in machine learning models for AMPylation site identification.

Main Methods:

• A new feature extraction pipeline was designed to encode key properties relevant to AMPylation.
• Machine learning classifiers were trained using numerical representations derived from the extracted features.
• Tenfold cross-validation was employed to assess model performance in distinguishing AMPylated from non-AMPylated sites.
• SHapley Additive exPlanations (SHAP) were used to analyze feature importance.

Main Results:

• The developed feature extraction framework demonstrated utility in machine learning models.
• The top-performing features achieved a Matthews Correlation Coefficient (MCC) score of 0.58, Accuracy of 0.8, AUC-ROC of 0.85, and F1 score of 0.73.
• Analysis elucidated model behavior based on monogram and bigram counts using SHAP.

Conclusions:

• The novel feature extraction pipeline significantly enhances the predictive performance of machine learning models for AMPylation sites.
• This approach offers a more efficient and accurate method for identifying AMPylation sites compared to experimental methods.
• The findings contribute to a better understanding of AMPylation and facilitate further research into its biological roles.