Poster #34: Tamar Peleg

A Computational Framework for Designing Engineered Targeting Proteins for Drug Delivery

Tamar Peleg1, Logan Hallee, Ph.D.2,3, Michael J. Donzanti, Ph.D.4, Jason P. Gleghorn, Ph.D.2,3,4 1Department of Medical and Molecular Sciences, University of Delaware, Newark, DE 2Center for Bioinformatics and Computational Biology, University of Delaware, Newark, DE 3Synthyra, PBLLC 4Department of Biomedical Engineering, University of Delaware, Newark, DE

Engineering ligands that reproduce natural receptor-mediated cell adhesion is valuable for drug delivery, immune modulation, diagnostics, and tissue engineering, but conventional screening to design and validate such mimetics is slow and expensive. We present a computational framework that accelerates protein binder design by coupling gradient-based sequence optimization with structure-based critic evaluation. As a scaffold, we selected Designed Ankyrin Repeat Proteins (DARPins) for their high-yield bacterial expression, thermal stability, and modular repeat architecture, and used the framework to design mimetics of the leukocyte adhesion receptor LFA-1 that reproduce its interaction with ICAM-1. Within each internal repeat, seven variable positions flanking fixed N- and C-cap sequences were treated as optimizable, allowing the number of repeats to be varied independently of sequence identity. In the design phase, sequence logits were iteratively updated using a structure predictor (ESMFold2) to score folding confidence (ipTM), interface contact and proximity, and specificity at each optimization step. A full-model critic phase then re-scored selected candidates using ipSAE, residue and atom contact pairs within 5Å, and minimum distance to the target epitope to confirm binding-site accuracy. We used random sequence sampling (10,000 variants per scaffold, 90% bootstrap confidence intervals) to estimate the best value for each DARPin template, and showed that our optimization scheme consistently finds the approximate minimum PPLL sequence across DARPin scaffolds spanning 1-4 internal repeats. Structural validation showed that a three-internal-repeat DARPin best reproduced native LFA-1–ICAM-1 contacts (minimum ipSAE=0.45, ipTM=0.88), outperforming shorter or longer variants. Together, these results indicate that efficient in silico discovery of receptor-mimetic binders is feasible, providing a generalizable computational pipeline for high-throughput engineering of adhesion-molecule mimetics before experimental validation.