Integrating Physics-Based Simulations and De Novo Protein Design to Engineer a Mesothelin-Targeting Scaffold

DIPC Seminars

Speaker

Angelo Spinello
University of Palermo

When

2026/06/29
10:00

Place

DIPC Josebe Olarra Seminar Room

Host

Stefano Scoditti

Add to calendar

iCal

Kimika Teorikoa Seminar

Mesothelin (MSLN) is a validated target for cancer therapy, yet antibody-based strategies often suffer from off-target interactions and poor tissue penetration. To overcome these limitations, a high-affinity non-immunoglobulin scaffold based on the human fibronectin (Fn3) was previously engineered. Here, we present a hybrid computational pipeline integrating AI-driven predictions with physics-based simulations to model the Fn3/MSLN complex. Initial consensus using canonical protein-protein docking algorithms produced different binding poses compared to the deep-learning-based AlphaFold 3 predictions. To address this discrepancy, we subjected the top-ranked models from both approaches to extensive molecular dynamics (MD) simulations. Crucially, experimental validation via yeast surface display, flow cytometry, and site-directed mutagenesis strictly corroborated the AlphaFold 3-derived complex model, highlighting the capability of recent deep-learning architectures to capture native-like topologies in challenging non-antibody interfaces. Finally, to expand the therapeutic toolkit against MSLN, de novo binder design campaigns are being performed. This integrated pipeline demonstrates how combining generative AI architectures with rigorous physics-based refinements could accelerate the precision engineering of next-generation oncology therapeutics.