AI Tool Predicts Bioactive Peptide Functions From Amino Acid Sequences With 93% Accuracy | Research

The takeaway

An ensemble deep learning model called DeepBP predicted ACE-inhibitory peptides with 92.6% balanced accuracy and anticancer peptides with 77.9% accuracy, outperforming existing methods.

92.6% accuracy
for predicting blood pressure-lowering ACE-inhibitory peptides from amino acid sequences alone using the DeepBP AI model

What the researchers found

DeepBP achieved 92.6% balanced accuracy (MCC 0.831, AUC 0.966) for ACE inhibitory peptide prediction and 77.9% accuracy (MCC 0.558) for anticancer peptide prediction, surpassing existing computational methods.

Why it matters

Drug discovery typically screens thousands of peptides experimentally. An AI tool that accurately predicts bioactive properties from sequence alone could accelerate peptide drug and functional food ingredient development by dramatically reducing the number of candidates that need laboratory testing.

The numbers in context

DeepBP uses an ensemble approach combining multiple deep learning models for improved prediction accuracy across bioactive peptide categories.

How the study worked

Ensemble deep learning using CapsuleGAN, GRU, and CNN as base classifiers with weighted voting. Features extracted using ESM-2 protein language model. Trained and validated on ACE inhibitory peptide and anticancer peptide datasets. Source code publicly available on GitHub.

Who was studied

Computational analysis of bioactive peptide sequence datasets

What this study cannot tell us

AI predictions require experimental validation — high computational accuracy does not guarantee real-world bioactivity. Training data may not cover all bioactive peptide types. Performance on rare or novel peptide activities is unknown. The anticancer peptide prediction accuracy (77.9%) leaves room for improvement.

How to read the evidence

Preliminary evidence: computational method demonstrating improved prediction accuracy on benchmark datasets. Requires experimental validation to confirm real-world utility.

When this study was published

Published in 2024 in BMC Bioinformatics. Uses current deep learning and protein language model technology.

The bigger picture

As peptide therapeutics gain importance, computational tools that predict bioactivity become essential for managing the vast sequence space. DeepBP represents the convergence of protein language models and ensemble deep learning — an approach likely to become standard in peptide drug discovery pipelines.

Questions still open

Can DeepBP be extended to predict other peptide functions like antimicrobial, anti-inflammatory, or neuroprotective activity?
How well do computationally predicted bioactive peptides perform when synthesized and tested in the laboratory?
Could this tool be integrated into automated peptide drug discovery pipelines?

Common questions

How does AI predict peptide functions?

DeepBP analyzes the amino acid sequence of a peptide using protein language models (similar to how ChatGPT processes text) and three types of neural networks. By learning patterns from thousands of known bioactive peptides, it predicts whether a new sequence will have specific properties like blood pressure-lowering or anticancer activity.

Can this tool discover new drugs?

It can accelerate drug discovery by identifying promising peptide sequences for laboratory testing. Instead of testing thousands of peptides blindly, researchers can use DeepBP to prioritize the most likely candidates, saving time and money in the drug development process.

Read the original research

DeepBP: Ensemble deep learning strategy for bioactive peptide prediction.

BMC bioinformatics, 25(1), 352

View on PubMed ↗Publisher’s paper ↗

Citation

Zhang, Ming; Zhou, Jianren; Wang, Xiaohua; Wang, Xun; Ge, Fang. (2024). DeepBP: Ensemble deep learning strategy for bioactive peptide prediction.. BMC bioinformatics, 25(1), 352. https://doi.org/10.1186/s12859-024-05974-5