PIP-EL: A New Ensemble Learning Method for Improved Proinflammatory Peptide Predictions.

Area of Science:

• Biochemistry and Molecular Biology
• Immunology
• Computational Biology

Background:

• Proinflammatory cytokines are crucial in host defense and inflammation.
• Proinflammatory inducing peptides (PIPs) have therapeutic potential as antineoplastics, antibacterials, and vaccines.
• Advancements in sequencing generate vast protein data, necessitating efficient identification of novel PIPs.

Purpose of the Study:

• To develop an automated computational method for accurate and rapid identification of novel proinflammatory inducing peptides (PIPs).
• To introduce PIP-EL, a predictor utilizing ensemble learning (EL) for PIP identification.

Main Methods:

• Implemented ensemble learning (EL) by fusing 50 independent random forest (RF) models.
• Utilized five distinct feature compositions: amino acid, dipeptide, composition-transition-distribution, physicochemical properties, and amino acid index.
• Applied random under-sampling to address an imbalanced benchmarking dataset, generating 10 balanced models per composition.

Main Results:

• PIP-EL achieved a Matthews' correlation coefficient (MCC) of 0.435 in 5-fold cross-validation, outperforming individual classifiers by ~2-5%.
• On an independent dataset, PIP-EL demonstrated superior performance with an MCC of 0.454 compared to existing and in-house methods.
• The predictor offers a user-friendly web server, freely accessible to researchers.

Conclusions:

• PIP-EL is an effective computational tool for predicting proinflammatory inducing peptides (PIPs).
• The predictor will significantly aid researchers in peptide therapeutics and immunotherapy.
• The ensemble learning approach enhances prediction accuracy for novel PIP identification.