PROJECT SUMMARY
With tunable binding affinity, solubility, and specificity characteristics, peptide inhibitors can interrupt
biological processes from cell signaling to viral infection vectors. Unfortunately, it is an unsolved challenge to
design a peptide to possess specifically sought interaction characteristics. Leveraging current capabilities in
structural bioinformatics, we aim to develop a general design platform for peptides that will bind appreciably only
to a specific binding site on one target protein. To this end, we rank order candidate peptides by employing a
combination of data-mining, molecular docking and molecular dynamics simulation in a serial-style pipeline. The
verification stage is to experimentally measure the binding characteristics of the top candidates. Successive
experiments will be performed as the ranked ordered list is traversed. As the results are compiled, supervised
machine learning will be iteratively applied to re-rank the candidate list to identify peptides with binding
characteristics that are sought. In this project, the p53 protein and its MDM2 and sirtuin binding partners serve
as a model system where a strategic set of systematic experiments will be performed. Importantly, because p53
is a critical hub protein in humans that modulates cellular function, transcription, and proliferation, there is
considerable published data that can be used for controls regarding this tumor suppressor, as well as its
biomedically important partners MDM2 and sirtuin. The format of the experimental design affords testing of a
multivariate binding objective involving more than one binding partner. Compared to rational design strategies
for small molecules, the potential for the development of a peptide-based lead compound is considerably higher.
An outcome of this project will be two separate public-domain software tools. The first, called PepStream, will
generate a candidate list of peptides ordered by propensity to bind to a target site using open repositories of
sequence and structural data. The second, is a supervised machine learning tool that is integrated with the
results from experimental measurements for successively re-ranking the candidate list to enhance success rates.
