A Hybrid Ensemble Model for Predicting TNF-α Inducing Peptides

Abstract

Renowned for its critical function in inflammation, tumor necrosis factor-alpha (TNFalpha) is an important modulator of the immune system. TNF-alpha has a wellestablished role in causing inflammation, yet in certain situations, it can exhibit surprising anti-inflammatory properties. This dual character emphasizes how crucial it is to identify the peptides generated by TNF-alpha. In order to overcome the drawbacks of manual identification, which is expensive, our work presents Stacking, an ensemble learning approach based on stacking that is intended for accurate and effective TNFalpha-inducing peptide identification. Ten amino acid composition-based feature extraction techniques are examined, including AAC, APAAC, CKSAAP, CTDC, CT raid, DPC, Moran, PAAC, and PsecRAAC. With the help of a Logistic Regression meta-learner and five improved base learners (LGBM, RF, SVM, Decision Tree, and KNN), the stacking model achieves an excellent 91.50% identification rate, 0.8291 MCC, and 0.9180 specificity. We examine the effects of different enhancement strategies on the accuracy of TNF-alpha predictions through experimental assessments, contrasting single models with ensemble combinations based on stacking. Our findings show that the suggested model is far more effective than its individual equivalents in identifying peptides that induce TNF-alpha. Improved TNF-alpha prediction advances the search for anti-inflammatory drugs.