|Category||Computational Biology, Digital Health|
|Keywords||Antibody epitope mapping, antibody-antigen complexes, structure prediction|
We address the problem of identification of antibody epitopes using AI. Epitope mapping is critical in antibody development process, where a few candidates are selected from a large pool based on their epitopes. Experimental methods for epitope mapping are not high throughput; therefore, an accurate computational approach will have a large impact in reducing the number of required experiments and the required timeframe.
A novel deep-learning method/software for antibody epitope identification through accurate structural modeling of antibody-antigen complexes:
- End-to-end high throughput deep learning method for accurate structure modeling of antibody-antigen complexes
- simultaneous antibody folding and antigen docking
- applicable to IgG antibodies and heavy chain only antibodies
We have designed a deep-learning model that when given an antigen structure and antibody sequence, produces accurate complex models, including sidechains. The network simultaneously folds the antibody (folding) and determines the orientation of the antigen with respect to the antibody (docking). The model is tailored specifically to take advantage of the properties of antibodies, such that achieving transformational invariance by antibody structure alignment. This enables achieving high accuracy without multiple sequence alignments.
The approach can be used in antibody discovery projects to accurately identify antibody epitopes. It is also applicable to nanobody discovery. Due to high speed and accuracy, the method can be applied to mining antigen-specific antibody repertoires identified from infected or vaccinated individuals. This application of the program enables mapping immunogenic epitopes and can help in vaccine design. The group is available to provide services for companies.