AI Deep Learning Discovers Potent Blood Pressure-Lowering Peptide From Quinoa Seeds | Research

The takeaway

A deep learning model combined with molecular docking identified NLFRP, a quinoa-derived peptide with potent ACE-inhibitory activity (IC50 = 3.33 μM), demonstrating AI's power to accelerate bioactive peptide discovery.

IC50 = 3.33 μM
The AI-discovered quinoa peptide NLFRP inhibited ACE with an IC50 of just 3.33 μM — a potent value that validates the deep learning approach and places this food-derived peptide among the more active natural ACE inhibitors reported.

What the researchers found

The Pre-GRU deep learning model combined with molecular docking successfully identified ACE-inhibitory peptides from quinoa:

- Model accuracy: R²train = 0.86-0.88, R²test = 0.53-0.55

- 4 out of 7 synthesized candidate peptides showed ACE inhibition activity (57% hit rate)

- NLFRP was the most potent: IC50 = 3.33 ± 0.19 μM

- Molecular docking revealed NLFRP forms tetrahedral coordination with ACE's zinc-binding HEXXH motif, explaining its strong binding

- Transfer learning from adjacent peptide lengths addressed the common problem of limited training data

- Balanced MSE loss function improved prediction accuracy despite imbalanced datasets

Why it matters

Traditional bioactive peptide discovery involves time-consuming trial-and-error experimentation. By using AI to predict which peptides from a food source will have therapeutic activity, researchers can focus synthesis and testing on the most promising candidates — in this case achieving a 57% hit rate (4 of 7). This approach could dramatically accelerate the discovery of food-derived peptides for managing hypertension and other conditions, making functional foods more scientifically targeted.

How the study worked

A pretrained Gated Recurrent Unit (GRU) deep learning model was developed using transfer learning — pretraining on known ACE-inhibitory peptides of various lengths, then fine-tuning on quinoa-specific data to overcome limited training samples. The model predicted IC50 values for candidate peptides from germinated quinoa seed protein. Top candidates were further evaluated by molecular docking to simulate ACE binding. Seven top-ranked peptides were synthesized and experimentally tested for ACE inhibition in vitro.

What this study cannot tell us

The model's test set accuracy (R²=0.53-0.55) is moderate, suggesting room for improvement. Only 7 peptides were synthesized and tested, limiting validation scope. All ACE inhibition data are in vitro — no cell-based, animal, or human blood pressure studies were conducted. The quinoa-derived peptides' stability during digestion and their oral bioavailability were not assessed. The IC50 value, while potent in vitro, does not guarantee in vivo blood pressure-lowering effects.

How to read the evidence

This is a computational and in vitro study combining AI prediction with biochemical validation. While the approach is innovative and the results are promising, no animal or human studies have been conducted. It represents early-stage bioactive peptide discovery research.

When this study was published

Published in 2025, this study is at the cutting edge of AI-driven peptide discovery, applying modern deep learning techniques (transfer learning, GRU networks) to food science — a rapidly emerging field.

The bigger picture

AI is transforming peptide discovery across all domains — from drug design to food science. This study demonstrates that deep learning can overcome a key bottleneck: limited experimental data for specific protein sources. By using transfer learning, the model leveraged knowledge from well-studied ACE-inhibitory peptides to make predictions about quinoa-derived peptides with minimal training data. The approach is generalizable to any food protein source, potentially unlocking bioactive peptides from underexplored crops.

Questions still open

Does NLFRP survive gastrointestinal digestion and maintain ACE-inhibitory activity when consumed as part of quinoa food products?
Could this AI-driven approach be applied to discover bioactive peptides for targets beyond ACE, such as DPP-4 (diabetes) or BACE (Alzheimer's)?
Would germinated quinoa products naturally contain enough NLFRP to have meaningful blood pressure-lowering effects?

Common questions

Can eating quinoa lower blood pressure?

This study identified peptides in germinated quinoa seeds that can inhibit ACE — the same enzyme targeted by blood pressure medications like lisinopril. However, the peptides were tested in a laboratory setting, not in humans. Whether eating quinoa provides enough of these specific peptides in the right form to meaningfully lower blood pressure remains unknown. The peptides may be present in very small amounts or may be broken down differently during normal digestion.

How does AI help discover health-promoting peptides?

Traditional peptide discovery involves extracting proteins from food, breaking them into pieces, and testing each one for activity — a slow, expensive process. The AI approach used here predicts which peptide sequences are most likely to inhibit ACE before any lab work begins. The deep learning model learned patterns from thousands of known ACE-inhibitory peptides and applied that knowledge to predict activity of quinoa peptides. This focused the lab work on just 7 top candidates, achieving a 57% success rate.

Read the original research

Integration of pre-trained GRU and molecular docking for virtual screening of quinoa seed derived ACE inhibitory peptides: An innovative prediction strategy.

Food chemistry, 492(Pt 3), 145591

View on PubMed ↗Publisher’s paper ↗

Citation

He, Yanan; Deng, Zhiyang; Lyu, Yujiao; Yan, Yan; Liu, Jun; Liu, Haijie. (2025). Integration of pre-trained GRU and molecular docking for virtual screening of quinoa seed derived ACE inhibitory peptides: An innovative prediction strategy.. Food chemistry, 492(Pt 3), 145591. https://doi.org/10.1016/j.foodchem.2025.145591