Antibody Structure Prediction¶
Antibody structure prediction is the task of predicting the structure of antibody variable domains (FV, including VH and VL). It is a key step in antibody design and optimization, as it provides the structural basis for downstream tasks such as antibody-antigen docking and binding affinity prediction/optimization.
Challenges¶
The CDR regions of antibodies are highly variable both in sequence and structure due to VDJ recombination and somatic hypermutation. Multiple sequence alignment (MSA) is not very helpful for modeling these regions.
Features¶
- State-of-the-art algorithm: Empowered by the latest geometric deep learning techniques, GeoAbFold achieves excellent performance, with an average RMSD lower than OmegaFold and AlphaFold2 on all FV regions.
- Cloud-based, mass-scale: Support batched prediction and relaxation of multiple sequences on our cloud platform.
- Graphical interface: No need to configure environments or write command lines. Just click and go! The predicted structure can be easily visualized in the Structure Viewer, with the b-factor indicating the confidence of the prediction.
Inputs¶
To submit an Antibody Structure Prediction job, open the Project Editor and select "Antibody Structure Prediction" from the "Structure Modeling" dropdown menu.
- Antibodies: A list of input antibodies. You can either enter the sequences in the input box or upload a FASTA file (by clicking the "
" button). Each antibody is predicted independently.- Antibody Name: Name of the antibody. Defaults to "Antibody". To change it, hover above the antibody name, and click on the "
" button. - Sequence: Sequence of an antibody. Each chain shall not exceed 200 residues.
- (Upload Sequence): Upload your own antibodies (a .FASTA file) by clicking the "" button. Each antibody must have 2 chains with labels sharing the same prefix and ending with "_H" and "_L" respectively. Above is an example.
- Antibody Name: Name of the antibody. Defaults to "Antibody". To change it, hover above the antibody name, and click on the "
- Job Name: Name of the job. Note that the job name must be unique within the project.
Model & Parameters¶
GeoAbFold, our proprietary antibody structure prediction model, is available for this job. Parameters of this model are as follows.
- # cycles: The number of recycles to run (0-3). Defaults to 3.
- Relax structure: Whether to relax the model-generated protein using Amber (default to false).
- seed: Random seed for structure initialization (default to a random integer).
Results¶
Click the job name in the Files & Jobs panel to view the job results.
The result summary is stored in a CSV file, which can be downloaded by clicking the "
" button to the top-right of the Summary Table.
The summary table contains the following columns:
- name: FASTA label, as in input.
- sequence: concatenated FASTA sequence.
- plddt: Predicted lDDT (local Distance Difference Test) score for the generated antibody. Higher is better. The per-residue lDDT scores are stored in the b-factor of the output .pdb files. You could view them by changing the color theme of the Cartoon representation (of the Polymer component) to "Atom Properties > Uncertainty/Disorder".
In the rightmost column, you could click on the "
" button to view the predicted structure.
If you enabled the "relax" option, you would find two files in the dropdown menu: one for the unrelaxed structure and one for the relaxed structure.