Aim

Designing peptides that can interact with specific immune targets is a challenging task with important applications in therapeutic development. Our aim is to evaluate generative AI methods that design new peptides for TCR-pMHC complexes, focusing on whether the generated molecules are structurally realistic and form meaningful interactions with the target.

We benchmark different models using sequence, structure, and interaction-based metrics to compare their performance under the same conditions. This helps us understand which approaches are most reliable, how well they generalize to new targets, and where current methods still need improvement. The project aims to contribute to more robust and trustworthy computational peptide design.