mHPpred: Accurate identification of peptide hormones using multi-view feature learning

Area of Science:

• Biochemistry and Molecular Biology
• Bioinformatics and Computational Biology
• Endocrinology

Background:

• Peptide hormones regulate crucial physiological processes like growth, development, and homeostasis.
• Experimental identification of peptide hormones is challenging due to low abundance, instability, and complexity.
• Computational methods, particularly machine learning, offer promising alternatives for peptide hormone prediction.

Purpose of the Study:

• To explore and evaluate different computational frameworks for accurate peptide hormone identification.
• To develop a superior computational model for predicting peptide hormones, enhancing their therapeutic applications.
• To introduce mHPpred, a novel meta-model for peptide hormone prediction.

Main Methods:

• Evaluated 26 feature descriptors using baseline models to identify seven with high predictive potential.
• Selected top 20 baseline models and integrated their predictions to train a meta-model.
• Developed a light gradient boosting-based meta-model, mHPpred, utilizing a multi-view feature learning strategy.

Main Results:

• The meta-approach framework proved most suitable for peptide hormone identification.
• mHPpred significantly outperformed the existing HOPPred method on benchmarking and independent datasets.
• mHPpred demonstrated superior performance compared to hybrid and integrative framework approaches.

Conclusions:

• The multi-view feature learning strategy in mHPpred effectively captures discriminative features for accurate peptide hormone prediction.
• mHPpred represents a significant advancement in computational tools for peptide hormone identification.
• The developed mHPpred model is publicly accessible for broader research use.