Competition values were prepared for cluster analysis and binning by capping all competition values between 0 and 1. Competition values between antibodies i and j were averaged with the competition value for j and i when both were available. Cluster4x (Ginn, 2020) was used to cluster antibodies into three distinct groups using single value decomposition on the matrix of competition values.

A surface of the receptor-binding domain was generated in PyMOL (The PyMOL Molecular Graphics System, Version 1.2r3pre, Schrödinger, LLC) from chain E of PDB code 6YLA. A mesh was generated and iteratively contracted and restrained to the surface of the RBD to provide a smoother surface on which to direct antibody refinement, reducing intricate surface features which could lead to unrealistic exploration of local minima.

In order to provide an objective position for those antibodies of known structure (FD5D (unpublished), EY6A (Zhou et al., 2020), S309 (Pinto et al., 2020) and mAb 40), to reflect the occluded region, all non-hydrogen antibody atoms were found within 20 Å of any RBD atom, and likewise all RBD atoms within 20 Å of an antibody atom. From each group, the atoms with the lowest sum-of-square-lengths from all other members were identified and the midpoint of these two atoms was locked to the nearest vertex on the mesh. Solvent molecules were ignored, but in the case of S309, the glycan cofactor was included in the set of antibody atoms.

On an evaluation of the target function, either all unique pairs of antibodies were considered (all-pairs), or only unique pairs where one of the antibodies was fixed (fixed-pairs), depending on the stage of the minimization protocol. Competition levels were estimated for each pair of antibodies as described by f(x) in Equation 1

where r is the working radius of the antibody, set to 11 Å, accounting for the approximate antibody radius. The distance between the pair of antibodies at a given evaluation of the function is given by d in Angstroms. The target function was the sum of squared differences between the competition estimation and the competition value from SPR data.

Minimization was carried out globally by 1000 macrocycles of Monte Carlo-esque sampling using LBFGS refinement. A random starting position for each antibody was generated by randomly assigning a starting vertex on the RBD mesh and the target function minimized for 20 cycles considering data points for pairs with at least one fixed antibody, followed by 40 cycles for all data points. Between each cycle, antibody positions were locked onto the nearest mesh vertex. Depending on the starting positions of antibodies, results were a mixture of well-refined and poorly refined solutions. Results were ordered in ascending target function scores. Positions of antibodies for each result was passed into cluster4x as dummy C-alpha positions (Ginn, 2020). A clear self-consistent solution was enriched in lower target function scores and separated using cluster4x for further analysis. The average position for each antibody was chosen as the sampled position which had the lowest average square distance to very other sampled position, and the RMSD calculated from all contributing antibody positions.

Note: The content above has been extracted from a research article, so it may not display correctly.



Q&A
Please log in to submit your questions online.
Your question will be posted on the Bio-101 website. We will send your questions to the authors of this protocol and Bio-protocol community members who are experienced with this method. you will be informed using the email address associated with your Bio-protocol account.



We use cookies on this site to enhance your user experience. By using our website, you are agreeing to allow the storage of cookies on your computer.