AI structural modeling of the viral protein shows that Maftivimab, the antibody WHO selected for the new Ebola outbreak, misses its target on its own and works only as part of the full Inmazeb triple cocktail
AI structural modeling of the viral protein shows that Maftivimab, the antibody WHO selected for the new Ebola outbreak, misses its target on its own and works only as part of the full Inmazeb triple cocktail
The hemorrhagic fever outbreak in the Democratic Republic of the Congo and Uganda is caused by Bundibugyo virus, a distant relative of Zaire ebolavirus. Bundibugyo had previously caused only two small outbreaks. An international team from Uganda, the DRC, and several other countries sequenced 44 viral genomes, including 12 from the 2026 outbreak, and in a paper published on 26 September 2026 used AI to model how approved and experimental Ebola antibodies dock onto the surface of this particular strain.
The FDA approved two Ebola antibody therapies in 2020 on the basis of trials against Zaire ebolavirus: mAb114 (Ebanga) and the triple-antibody cocktail Inmazeb. Both physically bind to the glycoprotein on the viral surface (the protein the virus uses as a key to enter human cells) and block it. Bundibugyo differs from Zaire by nearly 35% in the amino acid composition of this protein, and testing antibodies against the live virus is hindered by the scarcity of maximum-containment laboratories.
The authors therefore identified mutations in the genomes at positions where these antibodies bind and ran the structures through AlphaFold 3, DeepMind's AI system that predicts how two molecules dock with each other (for instance, a virus and an antibody). The Rosetta software then converted binding strength into a number, the binding energy. The method has a known limitation: across thousands of antibody-antigen pairs it identifies roughly half of confirmed contacts and sometimes classifies non-binding pairs as binding, so the authors validated it on a well-studied case. AlphaFold 3 predicted the structure of mAb114 on Zaire virus, and the result nearly perfectly matched the experimental crystal structure.
On Bundibugyo, the same model predicted a different outcome for mAb114: two substitutions in the protein, E112D and P116A, cause the antibody to dock away from its intended site. Rosetta's calculation estimated the loss of binding strength at approximately 54% for the first substitution and 34% for the second.
The result for Inmazeb was more complex. Of its three component antibodies, only Odesivimab docked onto the target when modeled individually; Atoltivimab and Maftivimab both missed. When all three were modeled together, as they exist in the actual drug, each landed in its correct position: neighboring antibodies hold one another in the required conformation. The effect is especially pronounced for Maftivimab. A glycan chain near its binding site throws off its alignment when the antibody is modeled alone but does not interfere when the full triple complex is present.
A third therapy, the experimental cocktail MBP134, targets the fusion machinery the virus uses to merge with the cell membrane. Any mutation in this region would cost the virus its ability to infect cells, so the region is nearly identical across all 44 sequenced genomes. The modeling predicts that both MBP134 components bind without obstruction, and for one of them this has already been confirmed by a crystal structure of the Bundibugyo protein.
The authors describe their model as a static snapshot of molecules, without the motion that occurs inside a living organism, and insist on validation through experiments with live virus.
In May 2026, WHO experts placed MBP134, remdesivir, and Maftivimab individually on the priority list for trials against Bundibugyo, without the full Inmazeb triple cocktail. The new modeling points in the opposite direction: Maftivimab is the weakest of the three on its own, yet the working triple combination was not included in the WHO priority list. MBP134 and remdesivir are already being administered to some patients in the PARTNERS trial, which compares treatment options in patients with confirmed Bundibugyo; the full Inmazeb is not yet on that list.