Title:	Development and comparison of de novo small binder design pipelines using RFdiffusion/AF_unmasked and BindCraft2
DNr:	Berzelius-2026-294
Project Type:	LiU Berzelius
Principal Investigator:	Marta Carroni <marta.carroni@scilifelab.se>
Affiliation:	Stockholms universitet
Duration:	2026-10-13 – 2027-05-01
Classification:	10610
Homepage:	https://www.scilifelab.se/researchers/marta-carroni/
Keywords:

Abstract

Recent advances in in silico protein structure prediction and design, recognized by the 2024 Nobel Prize in Chemistry, have opened new opportunities for the de novo design of small proteins that bind specific molecular targets. These advances are particularly relevant to the development of protein binders targeting defined epitopes on viral or cellular antigens. Such site-specific binders may offer new therapeutic strategies for cancer and autoimmune diseases, for example, by masking immunogenic epitopes and thereby modulating pathogenic immune responses. However, the efficient computational design and experimental validation of these molecules remain challenging. Generative machine-learning pipelines, such as RFdiffusion (Torres et al., 2023) and BindCraft (Pacesa et al., 2025), now facilitate the design of novel protein binders, including helical domains and antibody fragment antigen-binding (Fab) domains (Bennett et al., 2024) directed against selected epitopes. In our previous work, we demonstrated that antibody designs generated using RFdiffusion can be evaluated with AF_unmasked (Mirabello et al., 2024) by restricting structural modelling to regions relevant to the binding interface, potentially reducing computational costs. We have also explored alternative design pipelines, including RFantibody (Bennett et al., 2026). However, our preliminary investigations into the design of monoclonal antibodies targeting epitopes on human interferon and transferrin receptor indicated that useful designs of multiple complementarity-determining regions (CDRs) remains a substantial computational challenge, an even more a challenge for the experimental validation of the thousands of constructs designed. Consequently, the present study will focus on the design of small non-Fab, epitope-specific protein binders, which may offer a more tractable alternative in terms of computational design and validation as well as wet-lab production of binders. This computational study will complement our ongoing experimental investigations into the binding landscape of polyclonal antibodies implicated in the pathogenesis of multiple sclerosis. Specifically, we aim to design novel protein binders that interfere with antibody recognition of disease-relevant antigens, including human anoctamin-2, glial cell adhesion molecule (GlialCAM), αB-crystallin (CryAB), and myelin basic protein (Lum et al., 2026). These antigens are already under investigation in our laboratory, providing an established experimental framework for evaluating the designed binders. Promising candidates will undergo biochemical and structural characterization to assess their binding properties and validate the computational predictions. Biolayer interferometry (BLI) will be used to quantify binding affinity and evaluate competition with antibody binding, while cryo-electron microscopy (cryo-EM) will be employed, where feasible, to characterize in details the binding interfaces and compare the experimentally determined interaction modes with the computational models. Together, these approaches will establish a computational-to-experimental workflow for developing and validating small, epitope-specific protein binders against antigens relevant to multiple sclerosis pathogenesis.