Related Experiment Video
Updated: Jan 9, 2026
08:04
Overlapping Peptide Library to Map Qa-1 Epitopes in a Protein
Published on: December 20, 2017
7.3K
Accurate single-domain scaffolding of three nonoverlapping protein epitopes using deep learning
Karla M Castro1, Joseph L Watson2,3, Jue Wang2,3
1Institute of Bioengineering, École Polytechnique Fédérale de Lausanne (EPFL), Lausanne, Switzerland.
Nature Chemical Biology
|December 5, 2025
Summary
Deep learning enables the design of novel single-domain proteins capable of presenting multiple functional sites. This multiepitope protein design approach improves immune responses against respiratory syncytial virus.
Area of Science:
- Structural biology
- Protein engineering
- Immunology
Background:
- Traditional de novo protein design excels at scaffolding single functional motifs.
- Naturally occurring proteins often feature multiple functional sites, presenting a design challenge.
- Scaffolding multiple distinct functional sites within a single protein domain remains a significant hurdle.
Purpose of the Study:
- To develop a deep learning-based approach for simultaneously scaffolding multiple functional sites in a single-domain protein.
- To design small single-domain immunogens presenting three distinct respiratory syncytial virus motifs.
- To evaluate the structural accuracy and immunogenicity of the designed multiepitope proteins.
Main Methods:
- Utilized generative deep learning models for de novo protein design.
- Designed single-domain proteins under 130 residues incorporating three specific respiratory syncytial virus motifs.
- Employed X-ray crystallography for structural validation of motif presentation.
- Assessed immunogenicity through cross-reactive titers and neutralizing responses in animal models.
Main Results:
- Successfully designed small single-domain proteins with unusual folds, showing little global similarity to existing Protein Data Bank structures.
- X-ray crystallography confirmed accurate presentation of all three designed motifs on the protein surface.
- The multiepitope design elicited improved cross-reactive titers and neutralizing responses compared to single-epitope immunogens.
- Demonstrated the feasibility of presenting three distinct binding surfaces within a compact single-domain protein.
Conclusions:
- Generative deep learning is a powerful tool for tackling complex protein design challenges, including the simultaneous scaffolding of multiple functional sites.
- The designed multiepitope immunogens show promise for enhanced vaccine development against respiratory syncytial virus.
- This work advances the field of de novo protein design by enabling the creation of sophisticated, multi-functional proteins.
