Graph Neural Networks Outperform Traditional Models for Predicting Blood Pressure Peptides | Research

The takeaway

A graph convolutional network using molecular graph representations outperformed traditional machine learning for predicting ACE-inhibitory peptides of varying lengths.

GCN outperformed all others
Graph convolutional networks using molecular graph representations predicted ACE-inhibitory activity better than all traditional ML approaches

What the researchers found

A graph convolutional network using molecular graph representations outperformed traditional machine learning models for predicting ACE-inhibitory peptides of varying lengths.

Why it matters

ACE inhibitors are among the most prescribed blood pressure drugs. Finding new ACE-inhibiting peptides from natural sources could provide alternative or complementary treatments.

The numbers in context

Peptides containing 2-10 amino acids were represented as molecular graphs. The GCN model was compared against random forest, support vector machine, and other ML models.

How the study worked

Computational study comparing graph convolutional network models to traditional ML models for predicting ACE-inhibitory activity of peptides with 2-10 amino acids.

Who was studied

Computational analysis of peptide databases

What this study cannot tell us

Computational predictions need experimental validation. The model's accuracy may vary for peptides outside the training data range.

How to read the evidence

Rated preliminary: computational method comparison study. Predicted peptides need experimental validation.

When this study was published

Published in 2024. Graph neural networks are a cutting-edge approach in computational drug discovery.

The bigger picture

ACE inhibitors are among the most prescribed drugs. Finding new ACE-inhibiting peptides from food and natural sources could provide alternative blood pressure treatments with fewer side effects.

Questions still open

Can the GCN model be validated with experimental testing of predicted peptides?
Does this approach work for other bioactive peptide targets?

Common questions

What are ACE-inhibitory peptides?

Natural peptides (often from food proteins) that can lower blood pressure by inhibiting the same enzyme targeted by drugs like lisinopril and ramipril.

Why use graph neural networks for peptides?

Traditional methods struggle to capture the complex 3D relationships in peptide molecules. Graph networks represent atoms and bonds as connected networks, preserving structural information.

Read the original research

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

International journal of biological macromolecules, 282(Pt 5), 137060

View on PubMed ↗Publisher’s paper ↗

Citation

Rong, Yating; Feng, Baolong; Cai, Xiaoshuang; Song, Hongjie; Wang, Lili; Wang, Yehui; Yan, Xinxu; Sun, Yulin; Zhao, Jinyong; Li, Ping; Yang, Huihui; Wang, Yutang; Wang, Fengzhong. (2024). Predicting variable-length ACE inhibitory peptides based on graph convolutional network.. International journal of biological macromolecules, 282(Pt 5), 137060. https://doi.org/10.1016/j.ijbiomac.2024.137060