In Silico tool for predicting, designing and scanning IL-2 inducing peptides

Area of Science:

• Immunology
• Computational Biology
• Bioinformatics

Background:

• Interleukin-2 (IL-2) is vital for immune system regulation and is used in cancer immunotherapy.
• Predicting IL-2-inducing peptides is crucial for developing effective immunotherapies.
• Existing methods have limitations in precision or coverage.

Purpose of the Study:

• To develop and validate a computational method for predicting IL-2-inducing peptides.
• To compare alignment-based and artificial intelligence (AI) approaches for this prediction task.
• To create a user-friendly tool for the scientific community.

Main Methods:

• Trained and validated models on a dataset of 6,574 experimentally validated Major histocompatibility complex (MHC) binders.
• Developed alignment-based methods and various AI models, including machine learning (ML), deep learning (DL), and large language models (LLM).
• Constructed ensemble models combining AI and alignment-based approaches, and tested on additional datasets with MHC non-binders.

Main Results:

• AI models, particularly an Extra Tree-based model, showed promising performance with an AUC of 0.82.
• The best ensemble model, integrating Extra Tree and MERCI, achieved an AUC of 0.84 and MCC of 0.51 on the main dataset.
• Ensemble models achieved higher AUCs (0.9 and 0.8) on alternate datasets, demonstrating robustness.

Conclusions:

• AI and ensemble methods significantly improve the prediction of IL-2-inducing peptides.
• The developed tool, IL2pred, provides a valuable resource for predicting, scanning, and designing IL-2-inducing peptides.
• IL2pred is accessible via a web server, facilitating advancements in cancer immunotherapy research.