hladr4pred2

hladr4pred2 predicts peptide binding and non-binding for the MHC Class II allele HLA-DRB1*04:01 to identify allele-specific peptide binders.

Key Features:

• Large-scale training dataset: Trained on 12,676 binders and 12,676 non-binders for model development.
• Machine learning algorithms: Employs Support Vector Machines (SVM) and Artificial Neural Networks (ANN) for prediction modeling.
• Cross-validation and data split: Models were optimized using five-fold cross-validation on 80% of the dataset and evaluated on an independent 20% validation set.
• Feature types and performance: Uses composition and binary profile features with maximum AUCs of 0.90 and 0.87 respectively, and integration of BLAST with the composition-based model improved AUC from 0.90 to 0.93.
• Realistic dataset evaluation: Evaluation on a realistic set of 12,676 binders and 86,300 non-binders produced a maximum AUC of 0.99.
• Hybrid approaches and motif identification: Utilizes alignment-free and alignment-based hybrid approaches to identify motifs highly specific to HLA-DRB1*04:01 binders.

Scientific Applications:

• Allele-specific binder screening: Scanning peptide libraries to predict binders and non-binders for HLA-DRB1*04:01.
• Vaccine and immunotherapy design: Supporting development of targeted immunotherapies and vaccines by predicting allele-specific peptide binders.
• Disease association studies: Enabling identification of peptides relevant to diseases associated with HLA-DRB1*04:01, including sclerosis, arthritis, diabetes, and COVID-19.

Methodology:

Models were trained using SVM and ANN on composition and binary profile features with five-fold cross-validation on 80% of the data and independent evaluation on 20%; BLAST integration with the composition model and alignment-free/alignment-based hybrid approaches were applied; datasets included 12,676 binders with 12,676 non-binders for model development and evaluation against a realistic set of 12,676 binders and 86,300 non-binders.