pLM4ACE: A protein language model based predictor for antihypertensive peptide screening.

Area of Science:

• Biotechnology
• Computational Biology
• Pharmacology

Background:

• Angiotensin-I converting enzyme (ACE) is a key regulator of the renin-angiotensin system.
• ACE is a validated drug target for managing hypertension.
• Developing effective ACE inhibitors is crucial for cardiovascular health.

Purpose of the Study:

• To develop and validate a protein language model (pLM) for screening ACE inhibitory peptides.
• To leverage evolutionary scale modeling (ESM-2) embeddings for enhanced peptide activity prediction.
• To compare the performance of ESM-2 embeddings against traditional peptide embedding methods.

Main Methods:

• Utilized ESM-2 embeddings for a protein language model.
• Trained and evaluated 65 different classifiers on experimental data.
• Compared ESM-2 with 12 conventional peptide embedding techniques and 5 machine learning models.
• Assessed model performance using balanced accuracy (BACC), Matthews correlation coefficient (MCC), and area under the curve (AUC).

Main Results:

• Logistic regression combined with ESM-2 embeddings achieved the highest performance (BACC: 0.883 ± 0.017, MCC: 0.77 ± 0.032, AUC: 0.96 ± 0.009).
• ESM-2 embeddings demonstrated superior performance over 12 traditional embedding methods.
• Multilayer perceptron and support vector machine models also showed strong compatibility with ESM-2 embeddings.

Conclusions:

• ESM-2 embeddings provide a powerful tool for enhancing the prediction of ACE inhibitory peptide activity.
• The developed models offer a promising approach for discovering novel ACE inhibitors.
• A publicly accessible webserver is available for utilizing the top-performing models.