AagingPEPred: Anti-aging Peptide Prediction Server

Welcome to AagingPEPred

AagingPred is a machine learning-based tool designed to predict anti-aging peptides. It employs models trained on descriptors that can be calculated from amino acid sequences. The tool is capable of accepting peptide sequences in Fasta format, with lengths ranging from 5 to 50 amino acid residues.

For detailed instructions on how to utilize the tool, users are advised to refer to the Help page, which provides comprehensive information on the tool's usage, features, and functionalities.

To gain a deeper understanding of the methods and models employed by AagingPred, individuals can visit the Model page. This resource offers insights into the underlying techniques, algorithms, and data employed in the development of AagingPred's predictive models.

Specifications

About the model

AagingPEPred is a machine learning model for predicting peptide anti-aging potential. Users input peptides in fasta format (5-50 amino acids) to obtain predictions. Using machine learning techniques, AagingPEPred analyzes peptide characteristics to assess anti-aging potential, aiding researchers in exploring peptide properties for therapeutic interventions.

A predictive model for assessing peptide anti-aging potential requires balanced datasets. We sourced 216 positive peptides from AagingBase and generated an equivalent negative dataset of 216 peptides using Python's random library. This balanced dataset facilitates model development.

The positive and negative datasets underwent feature calculation using modlAMP and Pfeatures, generating 481 descriptors each. Feature selection was performed using RFE, a tool for machine learning and data mining. The top 20 ranked features, selected for their relevance in predicting anti-aging potential, were used for model training.

We developed a prediction model for anti-aging peptides using various AI/ML algorithms. The algorithms employed in this study included Support Vector Machine (SVM), Random Forest Classification (RFC), XG Boost (XGB), and Multilayer Perceptron (MLP). Among these algorithms, Random Forest Classification (RFC) demonstrated the highest accuracy for both the training and test datasets, achieving an accuracy of 80%. Based on this outcome, RFC was selected as the preferred model for further analysis, specifically in the context of AagingPEPred.