il13pred

il13pred predicts peptides that induce interleukin-13 (IL-13) to identify peptide sequences that modulate IL-13–mediated immune responses.

Key Features:

• Biological target: Interleukin-13 (IL-13), an immunoregulatory cytokine predominantly secreted by activated T-helper 2 cells and implicated in airway hyperresponsiveness, glycoprotein hypersecretion, goblet cell hyperplasia, inhibition of tumor immunosurveillance, and elevated levels in COVID-19 patients.
• Training dataset: 313 validated IL-13 inducing peptides and 2,908 non-inducing human peptides sourced from the Immune Epitope Database (IEDB).
• Feature extraction: 9,165 sequence-based features were generated using Pfeature.
• Feature selection: A linear support vector classifier with L1 penalty (SVC-L1) reduced features to 95 key features, which were ranked and the top 10 features were selected for model construction.
• Machine learning: Multiple machine learning techniques were evaluated and an XGBoost model provided the best performance.
• Model evaluation: Models were trained and tested using five-fold cross-validation and assessed on an independent validation dataset, with XGBoost achieving area under the curve (AUC) of 0.83 on training and 0.80 on independent validation.
• Viral variant analysis: Analysis indicated certain SARS-CoV-2 variants have a higher propensity to induce IL-13.

Scientific Applications:

• Therapeutic design: Identification of IL-13 inducing peptides to inform design of safer protein therapeutics by minimizing IL-13–mediated adverse effects.
• Immunology and allergy research: Characterization of peptide sequences that modulate IL-13 for studies of Th2-mediated responses and allergic disease mechanisms.
• Oncology: Investigation of peptides that influence IL-13–linked inhibition of tumor immunosurveillance to inform cancer immunology studies.
• Virology and pathogen analysis: Assessment of viral variant peptides, including SARS-CoV-2, for their propensity to induce IL-13 in host responses.

Methodology:

Peptide data were obtained from IEDB; 9,165 features were computed using Pfeature; SVC-L1 selected 95 features and ranked them with the top 10 used to train multiple machine learning models including XGBoost, which was evaluated via five-fold cross-validation and independent validation (AUC reported).