Abstract:
Protein-protein interactions form the basis of diverse processes in homeostasis and
disease. Consequently, protein binders that promote or antagonize these interactions
can serve as potent tools for both research and therapeutic purposes. While protein
design methods are rapidly advancing, the design of epitope-directed binders relying
on the target structure alone remains a formidable challenge. Existing de novo design
approaches yield high-affinity binders only through the experimental screening of
large libraries of designed candidates. The low success rates can be attributed to
the dimensionality of the simultaneous search for an optimal binder scaffold, pose,
and sequence. Moreover, the limitations in accurately estimating numerous factors
contributing to a binding event further complicate the scoring process. This work aims to
explore new design strategies to create on-demand protein binders in a resource-efficient
manner.
First, I evaluate the utility of a tiered approach that separates the docking task
from interface design to reduce complexity of the problem. The docking step uses
a novel surface fingerprinting method, which enables ultra-fast estimation of surface
complementarity and retrieves viable binder scaffolds from a protein structure database.
As proof-of-concept, I adopt this strategy to design binders targeting the vascular
endothelial growth factor (VEGF), a key angiogenic molecule implicated in pathogenesis
of various cancers. I experimentally characterize a small number of design candidates
and show that two of them have nanomolar affinity to VEGF, inhibit proliferation and
survival of VEGF-dependent cells, and finally have a VEGF-suppressing effect in vivo.
Second, I investigate the feasibility of tensorizing energy calculations for protein
design. The direct projection of atomic interaction fields in three-dimensional tensors
condenses energy evaluations into a single matrix operation, greatly simplifying the
computational load. Through retrospective validation, I demonstrate that the tensorized
framework outperforms other design engines in terms of speed and accuracy. For
prospective validation, I deploy this framework to design multi-specific binders against
ligands of the epidermal growth factor receptor (EGFR). The tested designs bind
strongly to their targets and inhibit EGFR activity in vitro and in vivo.
This work offers innovative solutions to protein docking and design problems.
Integrated into the design framework, these solutions can be used to rapidly create
protein binders against diverse targets through a single in silico round.