AntiMicrob

Description

This server is designed to predict potential antibacterial regions in proteins. The prediction principle is based on the results from our article "Comparison of deep learning models with simple method to assess the problem of antimicrobial peptides prediction" (forthcoming). We have shown that a simple method using only the amino acid composition of peptides works as well as neural networks. So, our method predicts with an accuracy of 77%, more complex methods are better by 3%, which is well within the error. In our method, each amino acid is assigned an antimicrobial activity score, which can be seen on the scale on the right. For peptides, we obtained the average antimicrobial activity and, based on this, predicted the antimicrobial activity of the entire peptide. In whole proteins, we are looking for short fragments that have antimicrobial activity. On this site, these regions are highlighted in dark green. We obtained a measure of the antimicrobial activity of each amino acid by comparing two sets of peptides. Peptides from one set had antimicrobial activity, while peptides from the second did not. So, having two sets of protein sequences, you can get a scale that best separates these two sets. This feature will be implemented on the website (address).

Scale