Computational methods for the structure-based design of protein-binding peptides
Name
swanson-swans-phd-bio-2023-thesis.pdf
Description
Thesis PDF
Size
27.25 MB
Format
Adobe PDF
Checksum (MD5)
c4cbecce1a3d5b98567e00de2638ded4
Author(s)
Swanson, Sebastian Robles
Advisor(s)
Keating, Amy E.
Date Issued
June 2023
Publisher
Massachusetts Institute of Technology
Abstract
The de novo design of peptides that bind to target proteins could enable binding to specific epitopes, inhibition of natural interactions, and targeted degradation of proteins. Despite advances in protein engineering, this remains a challenging task due to the large space of peptide structures and inaccuracies in atomic energy functions. In this thesis, I introduce new computational methods for structure-based design using structural motifs from the Protein Data Bank (PDB). To sample peptide structures in the context of a target protein, I mine tertiary motifs from known structures in the PDB to identify surface-complementing fragments or “seeds”. I show that TERM-based seeds can describe known binding structures with high resolution: the vast majority of peptide binders from a non-redundant set of 486 peptide-protein complexes can be covered by seeds. Furthermore, I demonstrate that known peptide structures can be reconstructed with high accuracy from peptide-covering seeds. I develop two methods for combining seeds to sample larger peptide backbone structures. The first method combines seeds that satisfy geometric overlap criteria and the second method identifies loop fragments from the PDB to join spatially proximal seeds. To score peptide structures, I develop statistical potentials that capture distinct features of their interface structures: sequence-structure compatibility and designability. Through a series of computational benchmarks, I show that the statistical potentials can be used to identify seeds predicted to form favorable interface structures. As proof of concept, I use the methods to design peptide binders of multiple target proteins, some of which have no known peptide binder. The designs are structurally diverse and have Rosetta energies that are comparable to natural peptides. For some of the peptides, I show that AlphaFold can accurately predict the designed structure. Altogether, this work demonstrates the potential of applying structural motifs to the design of protein-binding peptides and highlights important directions for future work.
MIT Department
Massachusetts Institute of Technology. Department of Biology
Terms of Use
In Copyright - Educational Use Permitted
Copyright retained by author(s)
Persistent DSpace Link
