hoppred

hoppred predicts peptide hormones to identify hormone peptides for studies of hormone function, regulation, and their roles in health and disease.

Key Features:

• Dataset Utilization: Trained and evaluated on a balanced dataset of 1174 hormonal and 1174 non-hormonal peptide sequences.
• Similarity-Based Methods: Uses BLAST and MERCI for similarity-based prediction, which provide high-confidence matches but can yield no hits for some sequences.
• Machine Learning Models: Implements machine learning models including logistic regression, with logistic regression achieving AUROC 0.93 and accuracy 86% on an independent validation dataset.
• Ensemble Method: Combines similarity-based methods and machine learning in an ensemble, achieving AUROC 0.96, accuracy 89.79%, and Matthews correlation coefficient (MCC) 0.8 on the validation set.
• Hormone-Associated Motif Identification: Identifies hormone-associated motifs within peptide sequences to aid structural characterization of hormonal peptides.

Scientific Applications:

• Peptide hormone prediction and design: Enables prediction and design of hormone peptides for downstream experimental or computational studies.
• Studies of hormone function and regulation: Supports investigations into hormone function, regulation, and their roles in health and disease.
• Endocrinology, molecular biology, and bioinformatics research: Provides computational predictions and motif information for researchers in endocrinology, molecular biology, and bioinformatics.

Methodology:

Training and evaluation on a balanced dataset of 1174 hormonal and 1174 non-hormonal peptides; similarity searches with BLAST and MERCI; machine learning models including logistic regression; ensemble combination of similarity-based and machine learning approaches; motif identification; evaluation using AUROC, accuracy, and MCC on independent validation datasets.