Saturday, January 27, 2007

How Peptides bind Proteins

Peptides are short proteins (polypeptides) -- up to about 50 amino acids. They can convey signals and mediate some of the activities of modular proteins. For example, peptides regulate behavior (neuropeptides), mood (substance P), and digestion (glucagon) The rules for peptides binding to proteins are not established, so Hertz and Yanover developed a new “framework” called PepDist for analyzing these interactions computationally. They say that most previous methods used only “binary” (binding/not-binding) classifiers, but a better method would rank peptides and the best would predict affinity. PepDist learns peptide-peptide distance functions and uses these to predict affinity. They trained their model peptides bound to histocompatibility proteins, which constitute a large and diverse family. They claim that PepDist outperforms the (formerly) state of the art models, including SVMHC, NetMHC (these links work, theirs don't!), and RANKPEP. The best part is that you can easily test it yourself – they provide an on-line service.
Hertz T and Yanover C, "PepDist: a new framework for protein-peptide binding prediction based on learning peptide distance functions." BMC Bioinformatics. 2006 Mar 20;7 Suppl 1:S3.


Reuel said...

MHC is the wrong "teacher" for their slow learner

It seems a little strange that they tested the model on peptides binding to histocompatibility proteins, since these interactions are probably subject to special constraints, such as forming an integral part of the protein and ensuring that certain amino acid side chains are exposed for T cell recognition. Moreover, unlike typical peptide-protein interactions, peptides bind partially unfolded MHC molecules and the binding to MHC class I is practically irreversible.

They should test their model with free peptides binding reversibly to folded, mature proteins.

Anonymous said...

Multiple Comparisons

You can compare the results of many different peptide prediction methods from a page at the Immune Epitope site (predictions).

This allows you to set one peptide length and threshold.