Improved prediction of antibody V_L–V_Horientation

Abstract

Antibodies are important immune molecules with high commercial value and therapeutic interest because of their ability to bind diverse antigens. Computational prediction of antibody structure can quickly reveal valuable information about the nature of these antigen-binding interactions, but only if the models are of sufficient quality. To achieve high model quality during complementaritydetermining region (CDR) structural prediction, one must account for the V L -V H orientation. We developed a novel four-metric V L -V H orientation coordinate frame. Additionally, we extended the CDR grafting protocol in RosettaAntibody with a new method that diversifies V L -V H orientation by using 10 V L -V H orientation templates rather than a single one. We tested the multiple-template grafting protocol on two datasets of known antibody crystal structures. During the template-grafting phase, the new protocol improved the fraction of accurate V L -V H orientation predictions from only 26% (12/46) to 72% (33/46) of targets. After the full RosettaAntibody protocol, including CDR H3 remodeling and V L -V H re-orientation, the new protocol produced more candidate structures with accurate V L -V H orientation than the standard protocol in 43/46 targets (93%). The improved ability to predict V L -V H orientation will bolster predictions of other parts of the paratope, including the conformation of CDR H3, a grand challenge of antibody homology modeling.

Modeling and docking of antibody structures with Rosetta

“…In AMA-II, we found that non-native V L –V H orientations can lead to explicit interactions between the light chain and the CDR H3 loop that are indistinguishable from native interactions 5 . Using multiple V L –V H orientation templates 27 allows broader exploration of conformational space, sampling more low-scoring wells. Models generated from at least three different templates should be used to maximize the chance of capturing the native V L –V H orientation.…”

“…The two plots show distributions of the Heavy Opening Angle 27 as obtained by plot_LHOC.py for two different antibodies. The 10 distinct light-heavy orientation templates are represented by the circles.…”

“…In general, the V L –V H orientation is correctly captured by RosettaAntibody in 43 of 46 benchmark antibody targets 27 . On the other hand, the CDR H3 loop is modeled de novo , and loop model quality decreases with loop length.…”

“…Unlike the other segments, ten V L –V H templates are selected to mitigate the weak correlation between sequence and orientation 27 . Starting from the list of all possible templates ordered by bit score, the best match is selected as the first template.…”

“…Starting from the list of all possible templates ordered by bit score, the best match is selected as the first template. To diversify the initial V L –V H orientations, all templates with similar V L –V H orientations (0.5 OCD, see Marze & Gray 27 ) to this template are pruned from the list. The best match remaining in the list is selected as the second template, and candidate templates similar to the second template are now removed from the list.…”