To assess the state of the art in antibody 3D modeling, 11 unpublished high-resolution x-ray Fab crystal structures from diverse species and covering a wide range of antigen-binding site conformations were used as a benchmark to compare Fv models generated by seven structure prediction methodologies. The participants included: Accerlys Inc, Chemical Computer Group (CCG), Schrodinger, Jeff Gray's lab at John Hopkins University, Macromoltek, Astellas Pharma/Osaka University and Prediction of ImmunoGlobulin Structure (PIGS). The sequences of benchmark structures were submitted to the modelers and PIGS, and a set of models were generated for each structure. We provide here an overview of the organization, participants and main results of this second antibody modeling assessment (AMA-II). Also, we compare the results with the first antibody assessment published in this journal (Almagro et al., 2011;79:3050).

A new method for calculating the total conformational free energy of proteins in water solvent is presented. The method consists of a relatively brief simulation by molecular dynamics with explicit solvent (ES) molecules to produce a set of microstates of the macroscopic conformation. Conformational energy and entropy are obtained from the simulation, the latter in the quasi-harmonic approximation by analysis of the covariance matrix. The implicit solvent (IS) dielectric continuum model is used to calculate the average solvation free energy as the sum of the free energies of creating the solute-size hydrophobic cavity, of the van der Waals solute-solvent interactions, and of the polarization of water solvent by the solute's charges. The reliability of the solvation free energy depends on a number of factors: the details of arrangement of the protein's charges, especially those near the surface; the definition of the molecular surface; and the method chosen for solving the Poisson equation. Molecular dynamics simulation in explicit solvent relaxes the protein's conformation and allows polar surface groups to assume conformations compatible with interaction with solvent, while averaging of internal energy and solvation free energy tend to enhance the precision. Two recently developed methods--SIMS, for calculation of a smooth invariant molecular surface, and FAMBE, for solution of the Poisson equation via a fast adaptive multigrid boundary element--have been employed. The SIMS and FAMBE programs scale linearly with the number of atoms. SIMS is superior to Connolly's MS (molecular surface) program: it is faster, more accurate, and more stable, and it smooths singularities of the molecular surface. Solvation free energies calculated with these two programs do not depend on molecular position or orientation and are stable along a molecular dynamics trajectory. We have applied this method to calculate the conformational free energy of native and intentionally misfolded globular conformations of proteins (the EMBL set of deliberately misfolded proteins) and have obtained good discrimination in favor of the native conformations in all instances.

A blinded study to assess the state of the art in three-dimensional structure modeling of the variable region (Fv) of antibodies was conducted. Nine unpublished high-resolution x-ray Fab crystal structures covering a wide range of antigen-binding site conformations were used as benchmark to compare Fv models generated by four structure prediction methodologies. The methodologies included two homology modeling strategies independently developed by CCG (Chemical Computer Group) and Accerlys Inc, and two fully automated antibody modeling servers: PIGS (Prediction of ImmunoGlobulin Structure), based on the canonical structure model, and Rosetta Antibody Modeling, based on homology modeling and Rosetta structure prediction methodology. The benchmark structure sequences were submitted to Accelrys and CCG and a set of models for each of the nine antibody structures were generated. PIGS and Rosetta models were obtained using the default parameters of the servers. In most cases, we found good agreement between the models and x-ray structures. The average rmsd (root mean square deviation) values calculated over the backbone atoms between the models and structures were fairly consistent, around 1.2 Å. Average rmsd values of the framework and hypervariable loops with canonical structures (L1, L2, L3, H1, and H2) were close to 1.0 Å. H3 prediction yielded rmsd values around 3.0 Å for most of the models. Quality assessment of the models and the relative strengths and weaknesses of the methods are discussed. We hope this initiative will serve as a model of scientific partnership and look forward to future antibody modeling assessments.

To assess the state-of-the-art in antibody structure modeling, a blinded study was conducted. Eleven unpublished Fab crystal structures were used as a benchmark to compare Fv models generated by seven structure prediction methodologies. In the first round, each participant submitted three non-ranked complete Fv models for each target. In the second round, CDR-H3 modeling was performed in the context of the correct environment provided by the crystal structures with CDR-H3 removed. In this report we describe the reference structures and present our assessment of the models. Some of the essential sources of errors in the predictions were traced to the selection of the structure template, both in terms of the CDR canonical structures and VL/VH packing. On top of this, the errors present in the Protein Data Bank structures were sometimes propagated in the current models, which emphasized the need for the curated structural database devoid of errors. Modeling non-canonical structures, including CDR-H3, remains the biggest challenge for antibody structure prediction.

Studies of antibodies of known three-dimensional structure have revealed that insertion and deletion of amino acids at the hypervariable loops change the canonical structures, thus generating differences in the antigen-binding site topography. Such differences determine the size of the antigen with which the antibody interacts. Here, 59 unique antibodies determined at a resolution of 3.0 A or below, including 19 in complex with proteins, 18 with peptides and 22 with haptens, were analyzed to identify and characterize differences in the residues that are directly involved in the interaction with antigen, so-called specificity-determining residues (SDRs). It was found that antibodies use a similar number of SDRs to recognize proteins and peptides but contact haptens with five SDRs less. By using a score of SDR usage, differences in the location of the SDRs, depending on the type of antigen recognized, were then identified with precision. An analysis of the surface generated by the SDRs usage indicates that the differences found correlate well with the size of the antigen. Anti-protein antibodies have the largest SDR surface, with SDRs of high usage located in the edge of the surface. The SDR surface of anti-hapten antibodies is the smallest, with hot spots of contacts in the interior of the binding surface and buried in the V(L):V(H) interface. The SDR surface of anti-peptide antibodies has a size in between anti-protein and anti-hapten antibodies, with the SDRs of high usage located in the interior of the antigen-binding site but do not buried as in anti-hapten antibodies. These findings led to a fine-tuning of the model correlating differences in the antigen-binding site topography with its preference to recognize antigens of different size. Therefore, it is discussed how this knowledge should help to design antibody repertoires biased toward the recognition of antigens of predefined size.