AMYPred-FRL

AMYPred-FRL predicts amyloidogenic proteins using feature representation learning to identify proteins implicated in amyloidoses such as type 2 diabetes, Alzheimer's disease, and Parkinson's disease.

Key Features:

• Meta-Predictor Approach: Integrates six machine learning algorithms—extremely randomized trees, extreme gradient boosting, k-nearest neighbors, logistic regression, random forest, and support vector machines—within a meta-predictor framework.
• Feature Representation Learning: Utilizes ten distinct sequence-based feature descriptors to generate 60 probabilistic features (PFs).
• Optimization via LR-RFE: Applies logistic regression recursive feature elimination (LR-RFE) to select an optimal subset of 20 PFs from the initial 60.
• Hybrid Logistic Regression Model: Integrates the selected 20 PFs into a logistic regression–based hybrid model for final predictions.
• Predictive Performance: Evaluated by cross-validation and independent testing, reporting an accuracy of 0.873 and an MCC of 0.710, with improvements of 5.5% in accuracy and 16.1% in MCC over existing methods.

Scientific Applications:

• Biomarker Discovery: Supports identification of potential amyloid-related biomarkers for amyloidoses.
• Therapeutic Target Identification: Assists in identifying protein targets implicated in aggregation-related diseases such as type 2 diabetes, Alzheimer's disease, and Parkinson's disease.
• Protein Misfolding and Aggregation Studies: Provides computational predictions relevant to protein misfolding and insoluble fibril aggregate formation.
• Disease Mechanism Research: Supplies predictive evidence to support studies of pathogenesis in aggregation-related disorders.

Methodology:

Generates 60 probabilistic features from ten sequence-based descriptors, employs a meta-predictor integrating extremely randomized trees, extreme gradient boosting, k-nearest neighbors, logistic regression, random forest and support vector machines, uses logistic regression recursive feature elimination (LR-RFE) to select 20 PFs, constructs a logistic regression–based hybrid model, and evaluates performance by cross-validation and independent testing.