TIPred

TIPred predicts tyrosinase inhibitory peptides (TIPs) from peptide sequence information to accelerate discovery of candidates for studying and treating hyperpigmentation disorders related to melanin overproduction and tyrosinase instability.

Key Features:

• Stacked ensemble learning: Implements a stacking strategy that integrates multiple machine learning models into an ensemble.
• Baseline models: Trains and optimizes 130 baseline models derived from well-established machine learning algorithms to produce new probabilistic features.
• Heterogeneous feature encoding: Encodes input sequences using schemes that capture chemical structure properties, physicochemical attributes, and composition information.
• Probabilistic feature integration: Uses probabilistic outputs from baseline models as additional features for downstream prediction.
• Feature selection: Applies a feature selection approach to identify an optimal feature vector for model development.
• Evaluation protocols: Assesses predictive performance using tenfold cross-validation and independent test sets within the stacking framework.
• Performance metrics: Reported results include accuracy 0.923, Matthews correlation coefficient (MCC) 0.757, and area under the curve (AUC) 0.977.

Scientific Applications:

• TIP discovery: Prioritizes candidate tyrosinase inhibitory peptides for subsequent experimental validation.
• Hyperpigmentation research: Supports studies addressing hyperpigmentation disorders associated with melanin overproduction and tyrosinase activity instability.
• Melanin and tyrosinase studies: Facilitates investigation of melanin biosynthesis pathways and tyrosinase-related mechanisms.
• Basic and translational research: Assists basic research and potential clinical development of therapeutics targeting tyrosinase and pigmentation.

Methodology:

Constructs a stacked ensemble by training and optimizing 130 baseline models to generate probabilistic features, combines these with heterogeneous encodings of chemical structure, physicochemical attributes, and composition information, applies feature selection to obtain an optimal feature vector, and evaluates performance using tenfold cross-validation and independent tests within a stacking strategy.