Abstract
BACKGROUND: The current therapy for inflammatory and autoimmune disorders involves the use of nonspecific anti-inflammatory drugs and other immunosuppressant, which are often accompanied with potential side effects. As an alternative therapy, anti-inflammatory peptides are recently being exploited as anti-inflammatory agents for treatment of various inflammatory diseases such as Alzheimer's disease and rheumatoid arthritis. Thus, understanding the correlation between amino acid sequence and its potential anti-inflammatory property is of great importance for the discovery of novel and efficient anti-inflammatory peptide-based therapeutics. METHODS: In this study, we have developed a prediction tool for the classification of peptides as anti-inflammatory epitopes or non anti-inflammatory epitopes. The training was performed using experimentally validated epitopes obtained from Immune epitope database and analysis resource database. Different sequence-based features and their hybrids with motif information were employed for development of support vector machine-based machine learning models. Similarly, machine learning models were also constructed using random forest. RESULTS: The composition and terminal residue conservation analysis of peptides revealed the dominance of leucine, serine, tyrosine and arginine residues in anti-inflammatory epitopes as compared to non anti-inflammatory epitopes. Similarly, the anti-inflammatory epitopes specific motifs were found to be rich in hydrophobic and polar residues. The hybrid of tripeptide composition-based support vector machine model and motif yielded the best performance on 10-fold cross validation with an accuracy of 78.1% and MCC of 0.58. The same displayed an accuracy of 72% and MCC of 0.45 on validation dataset, rejecting any possibility of over-fitting. The tripeptide composition-based random forest model displayed an accuracy of 0.8 and MCC of 0.59 on 10-fold cross validation, however, the accuracy (0.68) and MCC (0.31) was lower as compared to support vector machine model on validation dataset. Thus, the support vector machine model is implemented as the default model and an additional option of using the random forest model is provided. CONCLUSION: The prediction models along with tools for epitope mapping and similarity search have been provided as a web server which is freely accessible at http://metagenomics.iiserb.ac.in/antiinflam/ .