IL2pred

IL2pred predicts interleukin-2 (IL-2) inducing peptides to identify peptide sequences that stimulate IL-2 production for immunotherapy research.

Key Features:

• Primary dataset: Uses 6,574 experimentally validated Major Histocompatibility Complex (MHC) binders comprising 3,429 IL-2 inducing peptides and 3,145 non-inducing peptides.
• Amino acid pattern analysis: Identified residues such as alanine and leucine that are enriched in IL-2 inducing peptides versus non-inducing peptides.
• Alignment-based methods: Employed alignment-based approaches that showed high precision but limited coverage.
• Artificial intelligence models: Implements machine learning, deep learning, and large language models, including an Extra Tree-based model trained on dipeptide composition and peptide length that achieved an AUC of 0.82.
• Ensemble modeling: Integrates alignment-based methods with AI techniques; an ensemble combining the Extra Tree model and MERCI achieved an AUC of 0.84 and an MCC of 0.51 on the main dataset.
• Alternate datasets: Constructed Alternate Dataset 1 (3,429 inducers + 3,429 non-inducing MHC non-binders) and Alternate Dataset 2 (3,429 inducers + 3,439 non-inducers comprising binders and non-binders).
• Performance on alternate datasets: Reported ensemble performance with AUCs of 0.90 and 0.80 and MCCs of 0.61 and 0.44 on Alternate Datasets 1 and 2, respectively.

Scientific Applications:

• Prediction: Predicts whether short peptide sequences induce IL-2 production.
• Sequence scanning: Scans protein or peptide sequences to locate potential IL-2 inducing regions.
• Peptide selection and design: Supports selection and in silico design of candidate IL-2 inducing peptides for immunotherapy research.

Methodology:

Used a dataset of 6,574 MHC binders; performed amino acid composition and pattern analyses; applied alignment-based methods; trained ML/DL/LLM models including an Extra Tree model on dipeptide composition and peptide length; integrated MERCI with the Extra Tree model in an ensemble and evaluated performance using AUC and MCC on main and alternate datasets.