algpred2

algpred2 predicts allergenic proteins and identifies immunoglobulin E (IgE)-binding epitopes and allergenic regions within protein sequences to support allergy research and safety assessment.

Key Features:

• Dataset Utilization: Trained on a dataset comprising 10,075 allergens and an equal number of non-allergens and incorporating 10,451 experimentally validated IgE epitopes.
• BLAST Search: Performs Basic Local Alignment Search Tool (BLAST) searches against the allergen/non-allergen dataset to detect sequence similarity to known allergens.
• IgE Epitope Mapping: Searches for experimentally validated IgE epitopes from the Immune Epitope Database (IEDB) within query protein sequences to identify potential antigenic regions.
• Motif-Based Prediction: Detects characteristic motifs associated with allergens using multiple EM for motif elicitation and motif alignment search tools.
• Machine Learning Models: Employs machine learning techniques trained on the curated dataset to generate predictive models for allergenicity.
• Ensemble Approach: Integrates prediction scores from BLAST, epitope mapping, motif-based prediction, and machine learning models to produce combined predictions.
• Validation and Performance: Models achieved an AUC of 0.98 and an MCC of 0.85, with training on 80% of data using 5-fold cross-validation and validation on a separate 20% dataset ensuring ≤40% sequence similarity between sets.

Scientific Applications:

• Allergen Identification: Predicts potential allergenic proteins and maps specific IgE-binding epitopes within protein sequences.
• Safety Assessment: Assists evaluation of allergenic risk for novel food proteins, therapeutics, and biologically relevant substances.
• Research and Development: Supports design of hypoallergenic variants by identifying critical allergenic regions for modification.

Methodology:

Computational steps explicitly include BLAST searches against the allergen/non-allergen dataset, searching for IEDB-derived IgE epitopes, motif detection using multiple EM for motif elicitation and motif alignment search tools, training machine learning models on a dataset of 10,075 allergens and 10,075 non-allergens with 5-fold cross-validation on 80% training data, validation on a separate 20% dataset with ≤40% inter-set sequence similarity, and combining individual method scores in an ensemble.