Reverse Vaccinology 3.0 Flashcards (IPD reading)

Andreano, McLellan & Rappuoli, Nat Rev Micro 2026 Comment. the mpox / OPG153 story. TRAP cards = where the obvious summary is wrong. last card is the figure to explain.

what is reverse vaccinology 3.0 in one sentence?

sort memory B cells from people who beat a virus without using a bait antigen, find the antibodies that neutralize, then let AlphaFold 3 tell you which viral protein they hit, so you never have to solve a structure to learn the target

who wrote the reverse vaccinology 3.0 piece and what kind of article is it?

Andreano, McLellan and Rappuoli, Nature Reviews Microbiology 24(5):305-307, March 2026. it is a three page Comment with 10 references, not a review and not primary research. Rappuoli wrote the original reverse vaccinology paper in 2000

what was reverse vaccinology 1.0?

Rappuoli 2000: start from the pathogen's genome instead of growing the bug, list every surface protein it could make, and pick vaccine candidates from sequence. applied to Neisseria meningitidis serogroup B it found the protective antigens within 5 years and led to that pathogen's first vaccine

what was reverse vaccinology 2.0?

from 2016: start from human immunology instead. isolate human monoclonal antibodies from convalescent or infected people, solve the antibody-antigen complex, define the protective epitope, then design an improved immunogen

what is the RSV example in the Comment?

after the neutralizing antibody D25 was found, researchers stabilized RSV F in its prefusion conformation within 3 years, and a vaccine followed about 10 years after that. it is the proof that epitope first design works

what bottleneck does RV3.0 claim to remove?

the structural one. in 2.0 you must physically solve an antigen-antibody complex to learn what an antibody targets, and an antibody against an unsuspected protein might never be traced at all. AI structure prediction replaces that step

the four stages of the RV3.0 pipeline

genome sequencing to list candidate surface antigens; antigen agnostic single cell sorting of memory B cells with direct neutralization screening; AlphaFold 3 target deconvolution from antibody plus antigen sequences; cryo-EM confirmation of the prediction and the epitope

what is antigen agnostic screening and why is it the key trick?

normally you fish for B cells with a labelled bait antigen, so you only find antibodies to proteins you already suspected. sorting blind and testing what the cells secrete against whole virus lets an antibody against a protein nobody considered surface

what was known about mpox targets before this work?

conventional approaches had identified 7 of more than 35 MPXV antigens (M1, H3, E8, A29, A30, A35, B6), and used monovalently each neutralizes only the mature or the enveloped virion form, which is why candidates combine multiple antigens

what did the mpox campaign actually find?

memory B cells from MPXV infected or MVA-BN vaccinated donors, screened against a mix containing both mature AND enveloped virions, gave 12 mAbs cross neutralizing different MPXV forms and clades plus vaccinia. verbatim, "mAbs with these properties had never been described before"

what did AlphaFold 3 contribute in the mpox case?

fed the antibody sequences plus MPXV surface antigen sequences, within 5 days it gave two high confidence predictions naming OPG153 as a previously unknown neutralizing target. cryo-EM then confirmed it

what is OPG153?

the newly identified mpox neutralizing target, equivalent to MPXV A28. OPG stands for Orthopoxvirus gene. it was not one of the seven previously known targets

what was the immunogenicity result for OPG153?

monovalent OPG153 plus adjuvant induced MPXV and vaccinia neutralizing titres in mice comparable to the live attenuated MVA-BN vaccine and to multivalent mRNA candidates. the Comment gives no titre numbers

the authors' own failure rate

25%. AlphaFold 3 predicted the target with high confidence for only two of the eight mAbs that recognized the same antigen. three of every four antibodies were uninformative

the second limitation the authors state

AI robustness depends on large high quality datasets. structure has the PDB with more than 200,000 structures, while systems vaccinology, innate immunity and adjuvant biology have nothing comparable, so AI's impact there remains more limited

TRAP: what does "years to days" actually mean here?

the 5 days is the AlphaFold 3 step only. it begins after donor recruitment, antigen agnostic sorting and neutralization screening have already produced 12 validated mAbs, which is months of wet lab the headline leaves out

TRAP: how much of the mpox result is really AI?

arguably the assay design found OPG153: they screened against a mix of mature and enveloped virions, which is why they got cross form antibodies nobody had described. the AI deconvolved the target afterward

what does the Comment say about nanoparticles?

that antigens can be displayed "on natural or artificial nanoparticles to improve valency and immunogenicity". that downstream step is literally the King lab's product, and the Comment cites no King lab work at all

which supporting studies does the Comment cite?

mosaic sarbecovirus nanoparticles eliciting antibodies against strains not displayed (Cohen et al., Science 2021, Bjorkman lab); a generative framework restoring COV2-2130 activity against Omicron (Desautels et al., Nature 2024); viral language models predicting escape (Hie et al., Science 2021); eOD-GT8 60-mer priming VRC01 class precursors in humans (Leggat et al., Science 2022)

what does the Comment say about generative design?

RFdiffusion and similar models can build de novo backbones for epitope grafting, antigen stabilization and new immunogen architectures, superseding hand crafted tricks like cavity filling residues, engineered disulfides, helix capping prolines and optimized oligomerization domains

what is germline targeting, as used here?

designing a priming immunogen that binds the rare naive B cells whose unmutated receptors could eventually become broadly neutralizing, then boosting with immunogens progressively closer to the real virus. the authors concede it has not yet produced an approved HIV vaccine

why are mature vs enveloped virions worth knowing for this paper?

poxviruses exist in two infectious forms with different outer surfaces, so an antibody against one form often misses the other. that is why the seven old antigens had to be combined, and why screening against both forms produced broader antibodies

what conflicts of interest sit behind this Comment?

two patent applications covering exactly the mpox antibodies and the OPG153 vaccine antigen being celebrated, plus Novartis and GSK equity for Rappuoli. every data point is imported, most from one study by the same three authors

what is the honest scope of the RV3.0 claim?

n = 1 pathogen, efficacy stops at mouse neutralizing titres, no challenge protection and no clinical data. the "bacterial and parasitic pathogens" hedge in the article attaches to generative antigen design, not to the discovery pipeline

the best line to take from this paper into a conversation

AI's impact stops where PDB scale data stops. we can design the particle, we cannot yet predict the immune response to it, and that gap is where an ML person could contribute

explain this figure: reverse vaccinology 3.0 fig 1

left column is the two older pipelines. reverse vaccinology 1.0 runs targeted pathogen to genome information to antigen identification. reverse vaccinology 2.0 runs donor enrolment and blood collection to PBMCs to neutralizing mAbs to structural analysis. both feed into a brain shaped AI block in the middle, which branches to three outputs: antigen discovery, in silico antigen optimization and germline targeting design for vaccines; in silico mAb maturation and design for monoclonals; and engineering the immune system for new therapies. the honest read is that AI replaces the structure solving step, not the wet lab in front of it