Predicting variable-length ACE inhibitory peptides based on graph convolutional network.

Area of Science:

• Biochemistry
• Computational Chemistry
• Bioinformatics

Background:

• Traditional molecular descriptors have limitations in fully characterizing complex peptide structures for predicting angiotensin I-converting enzyme (ACE) inhibitory activity.
• Accurate prediction of ACE inhibitory peptides is crucial for developing functional foods with antihypertensive properties.

Purpose of the Study:

• To introduce and evaluate molecular graphs and a graph convolutional network (GCN) model for enhanced peptide characterization and prediction of ACE inhibitory activity.
• To compare the performance of the GCN model against traditional machine learning models using molecular descriptors.

Main Methods:

• Peptides (2-10 amino acids) were represented as molecular graphs.
• A graph convolutional network (GCN) model was developed for predicting variable-length peptide activity.
• The GCN model's performance was benchmarked against Random Forest (RF), Support Vector Machine (SVM), and k-Nearest Neighbor (kNN) models.

Main Results:

• The GCN model achieved an accuracy of 0.78 in predicting ACE inhibitory potential, outperforming ML models based on molecular descriptors.
• The GCN model demonstrated over 98% accuracy on an independent test set, significantly exceeding existing methods.
• Synthesized peptides VAPE and AQQKEP exhibited potent ACE inhibitory activity with low IC50 values (2.25 ± 0.11 and 3.75 ± 0.17 μM, respectively).

Conclusions:

• Molecular graphs and GCN models offer a superior approach for characterizing peptides and predicting ACE inhibitory activity compared to traditional methods.
• The developed GCN model is a powerful tool for rapid screening and identification of ACE inhibitory peptides.
• This research holds promise for the development of novel antihypertensive components in functional foods.